<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Andrea Schiona</title>
    <description>The latest articles on DEV Community by Andrea Schiona (@andrea_schiona).</description>
    <link>https://dev.to/andrea_schiona</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4080456%2Fca6112eb-b54e-490c-9bb4-1b39d51f60d2.png</url>
      <title>DEV Community: Andrea Schiona</title>
      <link>https://dev.to/andrea_schiona</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/andrea_schiona"/>
    <language>en</language>
    <item>
      <title>Journey to Autonomous Development: Five Levels of AI in the SDLC</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 21:21:44 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/journey-to-autonomous-development-five-levels-of-ai-in-the-sdlc-28gc</link>
      <guid>https://dev.to/andrea_schiona/journey-to-autonomous-development-five-levels-of-ai-in-the-sdlc-28gc</guid>
      <description>&lt;h1&gt;
  
  
  Journey to Autonomous Development: Five Levels of AI in the SDLC
&lt;/h1&gt;

&lt;blockquote&gt;
&lt;p&gt;Come l’AI sta ridefinendo il ciclo di vita del software — un modello di maturità in 5 livelli, dal coding assistito allo sviluppo autonomo.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Artificial Intelligence is reshaping how we build software. In the past, software development was entirely driven by human effort. Today, AI “co-pilots” are increasingly embedded throughout the Software Development Lifecycle (SDLC), from planning to maintenance. This article introduces a five-level maturity model for AI adoption in software development and walks through a practical example of an advanced AI-augmented workflow centered on GitHub and tools like GitHub Copilot, OpenCode, and Spec Kit.&lt;/p&gt;

&lt;h2&gt;
  
  
  Levels of AI Adoption in Software Development
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Level 1 — Traditional (No AI):&lt;/strong&gt; All development tasks are performed manually. Developers gather requirements, write code, create tests, and manage deployments without any AI assistance.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Level 2 — AI-Supported:&lt;/strong&gt; AI provides basic assistance. Tools offer passive help such as code autocomplete or simple recommendations, but they don’t make decisions. The human developers are fully in control.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Level 3 — AI-Assisted:&lt;/strong&gt; AI takes on a more proactive role. It can generate code snippets, suggest fixes or test cases, and highlight potential issues. Developers still verify and decide what to accept, but AI actively contributes to the work.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Level 4 — AI-Native:&lt;/strong&gt; AI becomes a collaborative partner throughout the SDLC. The development process is redesigned to integrate AI tools at each step: from planning (where AI helps define specs and break down tasks) to coding (where AI generates substantial portions of code) to testing and operations (where AI automates quality assurance and monitoring). Humans focus on guidance and oversight, while the AI handles much of the heavy lifting.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Level 5 — AI-Autonomous:&lt;/strong&gt; The autonomous development stage. AI systems handle nearly all aspects of development — interpreting requirements, writing and testing code, deploying software, and even monitoring and self-healing in production. Human involvement is minimal, limited to high-level guidance and governance. This is a visionary end-state that few organizations have fully reached, but it serves as a guiding star for the future of AI-driven development.&lt;/p&gt;

&lt;p&gt;To better understand these levels, consider how different SDLC phases change as you move up the adoption scale:&lt;/p&gt;

&lt;h3&gt;
  
  
  Table 1 — AI Adoption Levels Across SDLC Phases
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Requirements:&lt;/strong&gt; from manual gathering (L1) to AI-assisted spec generation (L4+)
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Design:&lt;/strong&gt; from manual docs (L1) to AI-proposed architecture with human review (L4+)
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Implementation:&lt;/strong&gt; from pure manual coding (L1) to AI-generated code with human approval (L3-L4)
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing:&lt;/strong&gt; from manual test cases (L1) to AI-generated tests and automated exploratory testing (L4+)
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deployment:&lt;/strong&gt; from manual release (L1) to AI-managed canary and self-healing rollouts (L5)
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Maintenance:&lt;/strong&gt; from reactive firefighting (L1) to AI-driven dependency updates and incident response (L4+)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;At higher levels, AI takes over more responsibilities, but human oversight and governance remain crucial. The progression from Level 1 to Level 5 is not just about adding tools — it’s about reimagining development workflows to fully leverage AI. For example, in Level 4 and 5, teams adopt methodologies like spec-driven development to give AI clear guidance, ensuring that AI-generated outputs meet the project’s intent and standards.&lt;/p&gt;

&lt;p&gt;Few organizations today have reached Level 5 (AI-autonomous) in practice — this is a frontier that’s still being explored. However, many teams are operating around Level 3–4, where AI is deeply integrated but humans remain in the loop. Even at these levels, the productivity boost can be dramatic: recent industry studies indicate AI coding assistants can nearly double development speed on certain tasks, and in some projects AI now generates almost half of the code. On the flip side, higher AI involvement means teams must address new challenges, such as managing AI-generated technical debt and enforcing security on AI-produced code. It’s a balancing act — using AI to go faster, while maintaining the quality, security, and ethical standards through diligent human oversight.&lt;/p&gt;




&lt;h2&gt;
  
  
  An AI-Augmented SDLC in Practice (GitHub-Centric)
&lt;/h2&gt;

&lt;p&gt;How does an advanced AI-supported workflow actually look in action? Let’s walk through a practical scenario at roughly the “AI-Native” (Level 4) stage, using GitHub as the central hub:&lt;/p&gt;

&lt;h3&gt;
  
  
  Requirements &amp;amp; Specification
&lt;/h3&gt;

&lt;p&gt;The team begins with a high-level idea for a new feature. Instead of writing a traditional spec from scratch, they use an AI-assisted approach. For instance, a product manager opens a GitHub Issue or uses the Spec Kit toolkit to describe the feature in natural language. Spec Kit takes this input and generates a detailed specification for the feature. The spec might include user stories, acceptance criteria, and success metrics — all derived from the manager’s description by the AI. The team reviews this auto-generated spec, making edits where needed. By having the AI do the heavy drafting, they ensure all aspects of the feature are considered upfront. They also capture any constraints in the spec so the AI is aware of them from the start.&lt;/p&gt;

&lt;h3&gt;
  
  
  Planning &amp;amp; Design
&lt;/h3&gt;

&lt;p&gt;Next, the team moves to technical design. Spec Kit can continue to assist by producing a technical plan based on the approved spec. The developers provide high-level guidance. The AI then outlines an architecture: it suggests the necessary modules or microservices, database schema changes, and integration points with existing systems. It might even draft an architecture diagram. At this stage, everyone discusses the AI’s proposal. They might ask the AI to iterate and compare alternatives. Once satisfied, the team has a clear architectural plan, recorded in the repository.&lt;/p&gt;

&lt;h3&gt;
  
  
  Task Breakdown
&lt;/h3&gt;

&lt;p&gt;With a spec and plan in place, it’s time to break the work into tasks. Here the AI automatically generates a list of implementation tasks. Each task is a well-defined piece of functionality. These tasks can be automatically turned into issues on GitHub with labels, descriptions, and even acceptance criteria drawn from the spec.&lt;/p&gt;

&lt;h3&gt;
  
  
  Coding &amp;amp; Implementation
&lt;/h3&gt;

&lt;p&gt;Now development begins in earnest. Developers still write code, but much of it is produced with AI assistance:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;They use GitHub Copilot in their VS Code editors to generate code from comments or prompts.&lt;/li&gt;
&lt;li&gt;For larger or more complex tasks, the team can use OpenCode, an open-source AI coding agent that runs locally or in the terminal.&lt;/li&gt;
&lt;li&gt;Branching and pull requests: Each feature is developed on a Git branch. When the AI completes a task, it may even create a commit and open a Pull Request on GitHub.&lt;/li&gt;
&lt;li&gt;Code review with AI: Before merging, the team conducts code reviews. Developers can use GitHub Copilot’s Code Review features or a custom GitHub Action to automatically analyze the PR. The AI review bot scans the changes and leaves comments.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Testing &amp;amp; Quality Assurance
&lt;/h3&gt;

&lt;p&gt;Thanks to AI integration, testing largely goes hand-in-hand with coding:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;When a developer creates a pull request, GitHub Actions automatically kicks off the CI pipeline.&lt;/li&gt;
&lt;li&gt;Additional AI-driven steps can be included, such as an action that runs an AI-based test generator on changed code.&lt;/li&gt;
&lt;li&gt;The AI’s involvement in testing isn’t limited to unit tests. It can simulate integration tests or user journeys.&lt;/li&gt;
&lt;li&gt;The QA engineers focus on monitoring these AI-driven test processes.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Deployment &amp;amp; Operations
&lt;/h3&gt;

&lt;p&gt;After a PR is reviewed and all checks pass, the code is merged. Continuous Deployment is configured via GitHub:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Automated deployments: A GitHub Actions workflow might automatically deploy the new feature to a staging environment. If no issues are found, the same pipeline can promote the release to production.&lt;/li&gt;
&lt;li&gt;Infrastructure as code &amp;amp; config optimization: The AI can propose changes to configuration files or even automatically adjust cloud settings.&lt;/li&gt;
&lt;li&gt;Post-deployment monitoring with AI: AIOps tools integrated with the system keep watch. Modern monitoring services often have AI/ML-based anomaly detection built-in.&lt;/li&gt;
&lt;li&gt;Automated responses: If the issue is a known type, the AI could automatically take action: roll back, open a branch with a fix, or apply a patch.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Maintenance &amp;amp; Continuous Improvement
&lt;/h3&gt;

&lt;p&gt;After deployment, the focus shifts to maintenance:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Automated updates: GitHub Dependabot keeps dependencies up to date, automatically opening PRs.&lt;/li&gt;
&lt;li&gt;Codebase health: AI code analysis runs periodically to refactor and improve code quality.&lt;/li&gt;
&lt;li&gt;Knowledge management: Documentation is kept in sync by AI.&lt;/li&gt;
&lt;li&gt;Incident handling: If a production incident occurs, AI aids the on-call team with preliminary analysis and may even draft a fix as a PR.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This end-to-end scenario shows how AI tools interconnect on a GitHub-centric platform to accelerate and automate development. Importantly, it’s not all-or-nothing — teams can mix and match these capabilities as needed.&lt;/p&gt;




&lt;h2&gt;
  
  
  Recommended AI Tools Across the SDLC
&lt;/h2&gt;

&lt;p&gt;To concretely illustrate the tools in such a setup, here’s a summary of recommended AI tools and services across the SDLC, particularly those that integrate well with GitHub:&lt;/p&gt;

&lt;h3&gt;
  
  
  Table 2 — AI Tools for an Advanced GitHub-Centered SDLC
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Specification:&lt;/strong&gt; Spec Kit, GitHub Issues with AI templates&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Planning:&lt;/strong&gt; Spec Kit plan generation, AI architecture assistants&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Task Breakdown:&lt;/strong&gt; Spec Kit tasks, GitHub Projects with automation&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Coding:&lt;/strong&gt; GitHub Copilot, OpenCode, Copilot CLI&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Code Review:&lt;/strong&gt; GitHub Copilot Review, custom GitHub Actions&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing:&lt;/strong&gt; AI test generators in CI, AI exploratory testing tools&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deployment:&lt;/strong&gt; GitHub Actions, AI canary management, AIOps platforms&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Monitoring:&lt;/strong&gt; AI/ML anomaly detection, GitHub Issues from alerts&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Maintenance:&lt;/strong&gt; Dependabot, AI refactoring tools, AI documentation sync&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This toolchain centers on GitHub not just as a code host, but as the integration point for AI in development. Each tool either lives within GitHub or connects to it, ensuring a seamless flow of information.&lt;/p&gt;




&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;The evolution of AI adoption in software development can be seen as a journey through these five levels of autonomy. Most teams today find themselves gradually climbing this ladder — maybe you started with a little AI autocomplete in your IDE (Level 2), then moved to having AI generate code and tests (Level 3). Some organizations are rethinking their entire development approach to be “AI-first” (Level 4), using new frameworks and practices that treat AI as a collaborator. And while Level 5 (full autonomy) remains an aspirational goal, each step toward it can yield significant benefits: faster development cycles, fewer mundane tasks for humans, and more reliable software.&lt;/p&gt;

&lt;p&gt;Crucially, higher AI involvement does not eliminate the need for human developers — it transforms their role. In an AI-augmented SDLC, developers spend less time on boilerplate and repetitive tasks and more time on creative design, complex problem-solving, and guiding the AI. Quality assurance shifts from writing test cases by hand to supervising an army of synthetic testers. Operations teams evolve from manually firefighting to building smarter automation and overseeing self-healing systems. In essence, AI frees humans to focus on what humans do best, while machines handle the drudgery and can even tackle problems at superhuman speed.&lt;/p&gt;

&lt;p&gt;For those looking to embrace an AI-supported development workflow, the best approach is iterative: start by introducing AI in one phase, then expand to other phases as your confidence and capability grow. Foster a culture of “human+AI partnership”, where AI is a reliable assistant and team members are skilled at directing and verifying AI output. Implement governance to ensure that increased automation doesn’t compromise quality or transparency.&lt;/p&gt;

&lt;p&gt;The future of software development is undoubtedly intertwined with AI. By understanding the maturity model of AI adoption and leveraging platforms like GitHub with intelligent tools, development teams can get started.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Andrea Schiona — AI Breakfast / Software Architecture&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>sdlc</category>
      <category>softwaredevelopment</category>
      <category>autonomous</category>
    </item>
    <item>
      <title>AI and the Future of Work: Risks, Numbers, and New Careers — AI Breakfast Special</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 21:03:11 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/ai-and-the-future-of-work-risks-numbers-and-new-careers-ai-breakfast-special-4l0a</link>
      <guid>https://dev.to/andrea_schiona/ai-and-the-future-of-work-risks-numbers-and-new-careers-ai-breakfast-special-4l0a</guid>
      <description>&lt;h1&gt;
  
  
  AI and the Future of Work: Risks, Numbers, and New Careers
&lt;/h1&gt;

&lt;blockquote&gt;
&lt;p&gt;AI Breakfast Special — What the data really says about automation, job losses, and the roles being created right now.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The public debate on AI and work is polarized between two narratives. On one side, the jobs apocalypse: robots and algorithms replacing millions of people. On the other, pure technological optimism: AI as a purely job-creating force. The data from major international organizations tells a more nuanced — and more urgent — story. This is not mass replacement, but a &lt;strong&gt;uneven transformation&lt;/strong&gt;: job contents are changing, new roles are emerging, and inequalities are widening between those who adapt and those who are left behind.&lt;/p&gt;

&lt;p&gt;This article breaks down the numbers, the most exposed sectors, the new careers already visible in the market, and the policies needed to steer the transition, with a specific focus on Italy and Europe.&lt;/p&gt;




&lt;h2&gt;
  
  
  The Numbers: Job Losses vs New Jobs
&lt;/h2&gt;

&lt;h3&gt;
  
  
  WEF Future of Jobs Report 2025/2026
&lt;/h3&gt;

&lt;p&gt;The World Economic Forum estimates that by 2030, automation and AI adoption could make &lt;strong&gt;approximately 85 million jobs&lt;/strong&gt; obsolete globally. But the same report predicts the creation of &lt;strong&gt;approximately 170 million new jobs&lt;/strong&gt;, for a net positive estimated at &lt;strong&gt;+85 million&lt;/strong&gt;. The catch: this figure is aggregated. The geographic, sectoral, and skill-level distribution is highly uneven.&lt;/p&gt;

&lt;h3&gt;
  
  
  McKinsey Global Institute — Agents, robots, and us (Nov 2025)
&lt;/h3&gt;

&lt;p&gt;McKinsey corrects the doomsday narrative: this is not sudden mass layoffs, but a &lt;strong&gt;prolonged restructuring&lt;/strong&gt; of work content. About &lt;strong&gt;40% of tasks&lt;/strong&gt; currently performed by workers could be automated by 2030, but only a minority of these tasks correspond to entire professions destined to disappear. More common is the case of roles that become &lt;strong&gt;hybrid&lt;/strong&gt;: an administrative worker moving from data entry to AI workflow supervision, a lawyer integrating automated document review tools.&lt;/p&gt;

&lt;h3&gt;
  
  
  OECD Employment Outlook 2024
&lt;/h3&gt;

&lt;p&gt;The OECD stresses that impact varies drastically by sector and skill level. Workers with &lt;strong&gt;low qualifications&lt;/strong&gt; and repetitive tasks are the most exposed, while professions requiring high human interaction, creativity, and complex problem-solving show &lt;strong&gt;structural resilience&lt;/strong&gt;. The organization warns, however, that without reskilling policies the risk is increased wage and territorial inequality.&lt;/p&gt;

&lt;h3&gt;
  
  
  IMF World Economic Outlook — AI and work (Apr 2025)
&lt;/h3&gt;

&lt;p&gt;The International Monetary Fund estimates that AI could contribute to a &lt;strong&gt;2-3% increase&lt;/strong&gt; in global productivity in the coming years, but with asymmetric distributional effects. In countries with low digitalization — including Italy — delays in adoption and training widen the gap with more advanced economies.&lt;/p&gt;




&lt;h2&gt;
  
  
  The Most Exposed Sectors: Who Is at Risk
&lt;/h2&gt;

&lt;p&gt;Not all jobs are equally at risk. Automation hits hardest activities with these characteristics:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Structured repetition&lt;/strong&gt;: data entry, filing, classification&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Explicit, predictable rules&lt;/strong&gt;: standardized administrative procedures&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Simple textual interaction&lt;/strong&gt;: FAQ responses, first-line customer support, basic ticketing&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The sectors where impact will be fastest and deepest:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Administration and offices&lt;/strong&gt;: basic accounting, practice management, invoicing, document logistics&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Customer service and call centers&lt;/strong&gt;: chatbots and virtual assistants already handle first-line support today&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Traditional manufacturing&lt;/strong&gt;: collaborative robots and predictive AI reduce the need for operators on standardized lines&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Retail and entry-level logistics&lt;/strong&gt;: warehousing, deliveries, automatic checkout, inventory&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transport&lt;/strong&gt;: autonomous driving for long-haul routes and internal logistics&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Conversely, more resilient are professions requiring:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Emotional intelligence and human connection&lt;/strong&gt;: nurses, teachers, therapists, social workers&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Creativity and strategy&lt;/strong&gt;: designers, architects, researchers, managers&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Unstructured physical work&lt;/strong&gt;: plumbers, electricians, healthcare operators&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Complex ethical and legal decisions&lt;/strong&gt;: judges, lawyers, auditors&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  New Roles Born with AI
&lt;/h2&gt;

&lt;p&gt;The other side of the coin is the rapid emergence of professions that didn’t exist ten years ago. These roles are not marginal: they require hybrid skills, are well-paid, and in strong growth.&lt;/p&gt;

&lt;h3&gt;
  
  
  AI Trainer and Prompt Engineer
&lt;/h3&gt;

&lt;p&gt;These professionals train, calibrate, and optimize AI models for specific tasks. They don’t just write prompts: they design fine-tuning datasets, define behavioral policies, evaluate response quality, and manage alignment between model and business objectives. In Italy, demand still lags behind qualified supply, but is growing in finance, retail, and public administration.&lt;/p&gt;

&lt;h3&gt;
  
  
  AI Governance Officer and Compliance Specialist
&lt;/h3&gt;

&lt;p&gt;With the &lt;strong&gt;EU AI Act&lt;/strong&gt; in force since 2024, high-risk AI systems must comply with transparency, documentation, and human control obligations. This has created a new professional pipeline: the governance officer defines internal policies, the compliance specialist verifies legal conformity, the risk assessor maps algorithmic risks. By 2027, every company with more than 250 employees will likely need at least one dedicated figure.&lt;/p&gt;

&lt;h3&gt;
  
  
  MLOps Engineer
&lt;/h3&gt;

&lt;p&gt;Between data science and operations, the MLOps engineer builds and maintains the pipelines that bring AI models into production. Manages model versioning, performance monitoring, continuous updates, and integration with enterprise systems. It’s one of the most sought-after tech roles, with 70% growth in the last two years.&lt;/p&gt;

&lt;h3&gt;
  
  
  Data Curator and AI Content Specialist
&lt;/h3&gt;

&lt;p&gt;AI models feed on data. The data curator selects, cleans, labels, and organizes high-quality datasets. It’s a hybrid figure between data scientist and archivist, increasingly important in healthcare, finance, and media. The AI content specialist, meanwhile, produces and reviews AI-generated content, ensuring accuracy, brand tone, and regulatory compliance.&lt;/p&gt;

&lt;h3&gt;
  
  
  Red Team Specialist and AI Safety Researcher
&lt;/h3&gt;

&lt;p&gt;When AI systems become critical — in cybersecurity, medicine, mobility — professionals are needed to test them systematically. The red team specialist looks for vulnerabilities, bias, unexpected behaviors. The AI safety researcher works on alignment theory and practice, in contexts ranging from labs to regulators.&lt;/p&gt;

&lt;h3&gt;
  
  
  Emerging Hybrid Roles
&lt;/h3&gt;

&lt;p&gt;Beyond specialized roles, &lt;strong&gt;hybrid professions&lt;/strong&gt; are growing: a doctor using AI-assisted diagnostic tools, a lawyer integrating predictive jurisprudential analysis, a journalist verifying algorithm-generated content, a project manager orchestrating human teams and AI agents. The trend is not replacement, but &lt;strong&gt;integration&lt;/strong&gt;.&lt;/p&gt;




&lt;h2&gt;
  
  
  The Italy Case: Between Delay and Opportunity
&lt;/h2&gt;

&lt;p&gt;Italy starts from a disadvantaged position in the AI adoption race, but the PNRR offers unprecedented resources to close the gap.&lt;/p&gt;

&lt;h3&gt;
  
  
  Data and Analysis
&lt;/h3&gt;

&lt;p&gt;According to &lt;strong&gt;Banca d’Italia&lt;/strong&gt; in the 2024 Annual Report, adoption of advanced digital technologies in Italian productive fabric is still limited: less than 20% of enterprises use AI tools structurally. &lt;strong&gt;Unioncamere Excelsior 2025&lt;/strong&gt; confirms that among workers at risk of transformation are mainly service-sector employees and specialized workers in routine processes.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;CEDEFOP Skill Anticipation 2025&lt;/strong&gt; estimates that in Italy between &lt;strong&gt;3 and 5 million workers&lt;/strong&gt; could see their roles deeply transformed by 2030. The risk is not mass unemployment, but the &lt;strong&gt;“impiegificio”&lt;/strong&gt;: underqualified positions, stagnant wages, working poor. To avoid it, a large-scale reskilling strategy is needed.&lt;/p&gt;

&lt;h3&gt;
  
  
  The PNRR as a Lever
&lt;/h3&gt;

&lt;p&gt;The National Recovery and Resilience Plan allocates significant resources to digital training, AI research, and industrial transition. The problem is implementation speed: small and medium enterprises struggle to access calls, training paths are often misaligned with market needs. An ecosystem is needed where universities, research centers, companies, and regions work together.&lt;/p&gt;

&lt;h3&gt;
  
  
  Italy’s Cultural Gap
&lt;/h3&gt;

&lt;p&gt;Beyond technical skills, there’s a problem of &lt;strong&gt;trust and understanding&lt;/strong&gt; of AI. In Italy, around 60% of workers declare they don’t know AI tools, and only 15% use them regularly in their work. This generational and cultural divide is the real bottleneck: it’s not just about courses, but a change of mindset.&lt;/p&gt;




&lt;h2&gt;
  
  
  Europe and the EU AI Act: Rules That Create Jobs
&lt;/h2&gt;

&lt;p&gt;Europe has chosen regulation over laissez-faire. The &lt;strong&gt;EU AI Act&lt;/strong&gt; classifies AI systems by risk and imposes increasing obligations: transparency, documentation, human control, periodic audits.&lt;/p&gt;

&lt;h3&gt;
  
  
  New Regulatory Professions
&lt;/h3&gt;

&lt;p&gt;This regulation is generating demand for new professional figures:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;AI Compliance Officer&lt;/strong&gt;: verifies systems comply with regulations&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AI Auditor&lt;/strong&gt;: conducts independent evaluations of high-risk models&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ethics Officer&lt;/strong&gt;: oversees ethical and social impact of algorithms&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Human-in-the-Loop Designer&lt;/strong&gt;: designs interfaces and processes that maintain human control&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;By 2030, around &lt;strong&gt;500,000 new jobs&lt;/strong&gt; related to AI governance are estimated to emerge in Europe. In Italy, with well-oriented PNRR and active policies, we can capture a significant share of this demand.&lt;/p&gt;

&lt;h3&gt;
  
  
  Global Competition
&lt;/h3&gt;

&lt;p&gt;Europe is not alone in the AI race. The United States and China invest far more in research and adoption. The risk is that Italy and Europe become &lt;strong&gt;consumers of others’ technologies&lt;/strong&gt;, without developing their own supply chain. The alternative is investing in applied research, startups, and training, transforming regulation from a cost to a competitive advantage: trust in the European AI system can become an asset.&lt;/p&gt;




&lt;h2&gt;
  
  
  The Real Risk: Uneven Transition
&lt;/h2&gt;

&lt;p&gt;AI is not a uniform event. It hits some sectors first, then others; benefits some categories, penalizes others; helps some regions, leaves others behind. The greatest danger is not total unemployment, but &lt;strong&gt;polarization&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Territorial Inequality
&lt;/h3&gt;

&lt;p&gt;Northern Italy, with a more industrialized productive structure and more dynamic SMEs, is better positioned to adopt AI and attract investment. The South, with greater exposure to traditional sectors and less digital infrastructure, risks suffering the transition without benefiting. Without convergence policies, AI could widen the North-South gap instead of reducing it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Generational and Gender Inequality
&lt;/h3&gt;

&lt;p&gt;Workers over 50 with limited digital skills are the most vulnerable. Women, overrepresented in administrative and service sectors at risk of automation, could suffer a disproportionate impact. Meanwhile, new AI roles today are predominantly male and young: active inclusion policies are needed.&lt;/p&gt;

&lt;h3&gt;
  
  
  The “Impiegificio” Risk
&lt;/h3&gt;

&lt;p&gt;If the transition is not governed, the danger is not unemployment, but &lt;strong&gt;mass dequalification&lt;/strong&gt;: workers expelled from traditional roles ending up in precarious, low-skill, low-pay positions. AI then becomes a tool of wage compression, not growth. History teaches that technological revolutions create wealth, but only when accompanied by investment in human capital.&lt;/p&gt;




&lt;h2&gt;
  
  
  What to Do: Individual and Systemic Strategy
&lt;/h2&gt;

&lt;h3&gt;
  
  
  For Workers
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Continuous training&lt;/strong&gt;: a degree is not enough. Every 3-5 years, update your skills. Platforms like Coursera, Udacity, and PNRR initiatives offer recognized paths.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hybrid skills&lt;/strong&gt;: combine your domain expertise with AI knowledge. An accountant learning AI for audit, a marketer learning AI-assisted content generation, a teacher integrating AI tools into teaching.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Develop “soft AI literacy”&lt;/strong&gt;: you don’t need to be a data scientist, but you need to evaluate output, recognize bias, and manage human-machine interaction.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  For Companies
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Gradual and participatory adoption&lt;/strong&gt;: AI is not installed, it is integrated. Involving workers in designing new workflows reduces resistance to change and increases transformation value.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Internal reskilling plans&lt;/strong&gt;: companies that invest in requalifying personnel have higher AI adoption rates and lower turnover.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Governance before technology&lt;/strong&gt;: define policies, responsibilities, and controls before implementing high-risk AI systems.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  For Public Policy
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Recognized micro-credentials&lt;/strong&gt;: fast, flexible, interoperable AI skill certification systems across regions and countries.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Lifelong learning accounts&lt;/strong&gt;: a personal fund for every worker, financed by companies and the State, for continuous training.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Active labor policies&lt;/strong&gt;: transition support, income support during training, incentives for hiring in strategic sectors.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Research and innovation&lt;/strong&gt;: increase public investment in AI applied to traditional Italian sectors — precision agriculture, tourism, manufacturing, healthcare.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  The Bottom Line: Govern, Don’t Defend
&lt;/h2&gt;

&lt;p&gt;Artificial intelligence does not destroy work uniformly. It transforms it. The real risk is the &lt;strong&gt;uneven transition&lt;/strong&gt;: those who update their skills become more productive and better paid, those who don’t risk exclusion.&lt;/p&gt;

&lt;p&gt;The goal is not to defend against automation, but to &lt;strong&gt;govern it&lt;/strong&gt;. This requires three things: widespread skills, clear rules, and bold investment. Italy has the opportunity to use AI to close its structural gaps, but only if it faces the transition as a collective project, not as a sum of individual choices.&lt;/p&gt;

&lt;p&gt;The future of work is not human against machine. It is human &lt;strong&gt;with&lt;/strong&gt; the machine.&lt;/p&gt;




&lt;h2&gt;
  
  
  Sources
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;McKinsey Global Institute — Agents, robots, and us (Nov 2025)&lt;/li&gt;
&lt;li&gt;WEF Future of Jobs Report 2025/2026&lt;/li&gt;
&lt;li&gt;OECD Employment Outlook 2024&lt;/li&gt;
&lt;li&gt;IMF World Economic Outlook — AI and work (Apr 2025)&lt;/li&gt;
&lt;li&gt;Banca d’Italia Annual Report 2024&lt;/li&gt;
&lt;li&gt;Unioncamere Excelsior 2025&lt;/li&gt;
&lt;li&gt;CEDEFOP Skill Anticipation 2025&lt;/li&gt;
&lt;li&gt;EU AI Act (Regulation EU 2024/1689)&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;🎙 &lt;strong&gt;Listen to the special podcast&lt;/strong&gt; (~5 min): &lt;a href="https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3" rel="noopener noreferrer"&gt;https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📻 Spotify: &lt;a href="https://open.spotify.com/show/033CI05HOIL6xYV16e88wV" rel="noopener noreferrer"&gt;https://open.spotify.com/show/033CI05HOIL6xYV16e88wV&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📬 Subscribe to the newsletter: &lt;a href="https://buttondown.com/italian-ai-breakfast" rel="noopener noreferrer"&gt;https://buttondown.com/italian-ai-breakfast&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📱 Telegram channel: &lt;a href="https://t.me/ItalianAIBreakfast" rel="noopener noreferrer"&gt;https://t.me/ItalianAIBreakfast&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Andrea Schiona — AI Breakfast&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>futureofwork</category>
      <category>career</category>
      <category>technology</category>
    </item>
    <item>
      <title>From Chaos to Clarity: How We Transformed a Brownfield Multi-Repo Project with AI Agents, OpenCode &amp; SpecKit</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 20:54:20 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/from-chaos-to-clarity-how-we-transformed-a-brownfield-multi-repo-project-with-ai-agents-opencode-38ke</link>
      <guid>https://dev.to/andrea_schiona/from-chaos-to-clarity-how-we-transformed-a-brownfield-multi-repo-project-with-ai-agents-opencode-38ke</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;Come abbiamo trasformato un progetto brownfield multi-repo in un workflow di sviluppo agentico con OpenCode e SpecKit — senza forzare un monorepo.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Modernizzare un progetto software legacy (“brownfield”) con l’assistenza dell’AI è oggi un obiettivo raggiungibile — anche quando il codice è distribuito su più repository. In questo articolo, ti guido attraverso come abbiamo evoluto un progetto Git multi-repository classico in un workflow di sviluppo assistito da AI utilizzando &lt;strong&gt;OpenCode&lt;/strong&gt; e &lt;strong&gt;GitHub SpecKit&lt;/strong&gt;. Lungo il percorso, confronteremo questo approccio con altri assistenti di coding AI come &lt;strong&gt;GitHub Copilot&lt;/strong&gt;, &lt;strong&gt;Amazon CodeWhisperer&lt;/strong&gt; e &lt;strong&gt;Sourcegraph Cody&lt;/strong&gt;, e condivideremo lezioni pratiche apprese.&lt;/p&gt;

&lt;p&gt;I nostri obiettivi erano chiari:&lt;/p&gt;

&lt;p&gt;✅ Una vista unificata e coerente dell’applicazione multi-repo (frontend, backend, componenti condivisi)&lt;br&gt;
✅ Generazione e manutenzione automatiche della documentazione tecnica “as-is”&lt;br&gt;
✅ Un flusso operativo ripetibile per analisi, sviluppo di feature ed esecuzione locale&lt;/p&gt;

&lt;p&gt;Questo percorso si inserisce in una esplorazione più ampia che sto documentando. Se ti interessano le basi architetturali di questo approccio, ti consiglio di leggere il mio articolo precedente:&lt;br&gt;
👉 &lt;a href="https://medium.com/p/627795029809" rel="noopener noreferrer"&gt;https://medium.com/p/627795029809&lt;/a&gt;&lt;/p&gt;


&lt;h2&gt;
  
  
  🧩 La Sfida del Multi-Repo Brownfield
&lt;/h2&gt;

&lt;p&gt;In un setup enterprise tipico, una singola applicazione spesso si estende su più repository Git — ad esempio, una UI frontend, uno o più servizi backend, librerie condivise e codice di infrastruttura. Ogni repo ha la propria storia e il proprio ciclo di deployment. Questa separazione è intenzionale e vantaggiosa per l’autonomia dei team, ma crea problemi quando gli assistenti di coding AI devono comprendere il quadro completo.&lt;/p&gt;

&lt;p&gt;I progetti brownfield amplificano questa sfida: la codebase è grande, matura e plasmata da anni di decisioni e convenzioni. Introdurre un agente AI in questo ambiente significa che deve apprendere i pattern del progetto e non rompere nulla. L’AI necessita di una vista panoramica del sistema per essere veramente utile — altrimenti potrebbe suggerire modifiche che ignorano dipendenze in un altro repo o codice che non segue le convenzioni esistenti.&lt;/p&gt;

&lt;p&gt;Perché non passare direttamente a un monorepo? Migrare tutto il codice in un solo repository potrebbe dare all’AI il contesto completo, ma è di solito impraticabile per un sistema maturo. Il passaggio a un monorepo introduce overhead pesanti (cambiamenti CI/CD, tooling, interruzione dei flussi di team) e rischi significativi senza benefici immediati. A meno che non si parta da zero o i repository siano già strettamente accoppiati, una migrazione a monorepo raramente vale la pena per questo scopo. Invece, abbiamo cercato una soluzione non disruptiva: mantenere i repo separati ma lasciare che l’AI li veda e lavori su di essi come se fossero unificati.&lt;/p&gt;


&lt;h2&gt;
  
  
  🛠️ Implementazione Passo-Passo
&lt;/h2&gt;
&lt;h3&gt;
  
  
  Step 1: Consolida i Repository Localmente
&lt;/h3&gt;

&lt;p&gt;Abbiamo creato un workspace locale e clonato tutti i repo Git rilevanti (frontend, backend, componenti condivisi) in sottocartelle. Questo fornisce una struttura di file unificata per l’analisi.&lt;br&gt;
📎 Per approfondire questo setup e la sua rationale, vedi il mio articolo: &lt;a href="https://medium.com/p/627795029809" rel="noopener noreferrer"&gt;https://medium.com/p/627795029809&lt;/a&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;mkdir &lt;/span&gt;workspace
&lt;span class="nb"&gt;cd &lt;/span&gt;workspace
git clone &amp;lt;repo-frontend-url&amp;gt; frontend
git clone &amp;lt;repo-backend-url&amp;gt; backend
git clone &amp;lt;repo-shared-url&amp;gt; shared
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Step 2: Valida una Baseline Pulita
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git &lt;span class="nt"&gt;-C&lt;/span&gt; frontend status
git &lt;span class="nt"&gt;-C&lt;/span&gt; backend status
git &lt;span class="nt"&gt;-C&lt;/span&gt; shared status

git &lt;span class="nt"&gt;-C&lt;/span&gt; frontend remote &lt;span class="nt"&gt;-v&lt;/span&gt;
git &lt;span class="nt"&gt;-C&lt;/span&gt; backend remote &lt;span class="nt"&gt;-v&lt;/span&gt;
git &lt;span class="nt"&gt;-C&lt;/span&gt; shared remote &lt;span class="nt"&gt;-v&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Step 3: Crea un Repository Dedicato per le Specifiche
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;mkdir &lt;/span&gt;spec
&lt;span class="nb"&gt;cd &lt;/span&gt;spec
git init
git branch &lt;span class="nt"&gt;-M&lt;/span&gt; main
git remote add origin &amp;lt;repo-spec-url&amp;gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Step 4: Inizializza OpenCode e SpecKit
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;opencode init
speckit init
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Step 5: Genera la Documentazione AS-IS
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;speckit scan &lt;span class="nt"&gt;--source&lt;/span&gt; ../workspace
speckit generate as-is
speckit &lt;span class="nb"&gt;export&lt;/span&gt; &lt;span class="nt"&gt;--format&lt;/span&gt; md &lt;span class="nt"&gt;--out&lt;/span&gt; ./docs/as-is
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Step 6: Commit e Verifica
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git add &lt;span class="nb"&gt;.&lt;/span&gt;
git commit &lt;span class="nt"&gt;-m&lt;/span&gt; &lt;span class="s2"&gt;"chore(spec): initial AS-IS baseline"&lt;/span&gt;
git push &lt;span class="nt"&gt;-u&lt;/span&gt; origin main
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Poi verifica l’esecuzione locale:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;cd&lt;/span&gt; ../workspace/frontend
npm &lt;span class="nb"&gt;install
&lt;/span&gt;npm start
&lt;span class="nb"&gt;cd&lt;/span&gt; ../backend
npm &lt;span class="nb"&gt;install
&lt;/span&gt;npm run dev
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;h2&gt;
  
  
  🤖 Usare Agenti AI su un Sistema Brownfield
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Comprensione Cross-Repo
&lt;/h3&gt;

&lt;p&gt;Con tutti i repository caricati nel contesto, l’AI poteva ragionare su modifiche che coinvolgevano frontend, backend e codice condiviso. Identificava correttamente dove posizionare le modifiche e ci chiedeva di confermare quale repo modificare.&lt;/p&gt;

&lt;h3&gt;
  
  
  Aggiornamenti Automatici della Documentazione
&lt;/h3&gt;

&lt;p&gt;Dopo aver implementato le modifiche, riavviavamo:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;speckit scan &lt;span class="nt"&gt;--source&lt;/span&gt; ../workspace
speckit generate as-is
speckit &lt;span class="nb"&gt;export&lt;/span&gt; &lt;span class="nt"&gt;--format&lt;/span&gt; md &lt;span class="nt"&gt;--out&lt;/span&gt; ./docs/as-is
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  Sviluppo Iterativo, Spec-First
&lt;/h3&gt;

&lt;p&gt;Abbiamo seguito la metodologia di SpecKit:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Scrivi o aggiorna una specifica&lt;/li&gt;
&lt;li&gt;Genera un piano con &lt;code&gt;/plan&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Scomponi i task con &lt;code&gt;/tasks&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Implementa con OpenCode&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Guidare l’AI con le Regole
&lt;/h3&gt;

&lt;p&gt;Abbiamo perfezionato la nostra Costituzione per:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Prevenire la duplicazione del codice&lt;/li&gt;
&lt;li&gt;Applicare il layering architetturale&lt;/li&gt;
&lt;li&gt;Vietare anti-pattern (es. try/catch nei componenti UI)
📎 Ho esplorato questo argomento in modo più approfondito in un articolo dedicato: &lt;a href="https://medium.com/p/da4204b3286a" rel="noopener noreferrer"&gt;https://medium.com/p/da4204b3286a&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  📚 Perché Tenere le Specifiche e la Documentazione in un Repository Separato?
&lt;/h2&gt;

&lt;p&gt;Abbiamo scelto di memorizzare tutte le specifiche e la documentazione in un repository Git dedicato (&lt;code&gt;spec&lt;/code&gt;) invece di mescolarle nel codice dell’applicazione. Questo ha permesso:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Versioning e storia indipendenti&lt;/li&gt;
&lt;li&gt;Workflow di review dedicati&lt;/li&gt;
&lt;li&gt;Meno rumore nei repository del codice&lt;/li&gt;
&lt;li&gt;Migliore governance della documentazione cross-repo&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Alternative come incorporare le specifiche in ogni repo o memorizzarle in una cartella non versionata sono state scartate a causa della frammentazione, della mancanza di versioning e della scarsa revisionabilità.&lt;/p&gt;




&lt;h2&gt;
  
  
  ⚖️ Confronto tra Tool: Assistenti AI per Progetti Brownfield
&lt;/h2&gt;

&lt;p&gt;Un aspetto chiave della nostra valutazione è stato il confronto tra diversi strumenti di AI assistance. &lt;strong&gt;GitHub Copilot&lt;/strong&gt; eccelle nel suggerire codice inline all’interno dell’IDE, ma ha un contesto limitato al file corrente o alla piccola porzione di codice aperta. &lt;strong&gt;Amazon CodeWhisperer&lt;/strong&gt; e &lt;strong&gt;Sourcegraph Cody&lt;/strong&gt; offrono una comprensione più ampia del repository, ma faticano ancora a coordinare modifiche coerenti su più repo separati. &lt;strong&gt;OpenCode&lt;/strong&gt;, combinato con &lt;strong&gt;SpecKit&lt;/strong&gt;, si è distinto per la capacità di operare su un workspace consolidato, applicare regole architetturali e mantenere la documentazione sincronizzata — senza richiedere una migrazione a monorepo.&lt;/p&gt;

&lt;p&gt;Per progetti brownfield complessi, la scelta dello strumento non è solo una questione di qualità del completamento codice, ma di &lt;strong&gt;governance del cambiamento&lt;/strong&gt;: quanto l’AI riesce a capire le regole del tuo sistema e a rispettarle durante modifiche estese.&lt;/p&gt;




&lt;h2&gt;
  
  
  🚀 Conclusione
&lt;/h2&gt;

&lt;p&gt;Trasformare un progetto brownfield multi-repo in un workflow assistito da AI non è solo possibile — è potente. Federando i nostri repository in un singolo contesto AI e adottando un processo di sviluppo spec-first, abbiamo ottenuto:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Coerenza tra i repository&lt;/li&gt;
&lt;li&gt;Documentazione sempre aggiornata&lt;/li&gt;
&lt;li&gt;Iterazioni più veloci con meno overhead manuale&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Non abbiamo dovuto ristrutturare il progetto o forzare un monorepo. Invece, abbiamo introdotto tooling AI in modo complementare ai flussi esistenti. Ad ogni iterazione, l’AI si è allineata maggiormente alla nostra architettura e alle nostre convenzioni.&lt;/p&gt;

&lt;p&gt;Questo approccio è adatto a tutti? Se gestisci una codebase grande e complessa distribuita su più repository, credo che meriti seria considerazione. Con il setup e la governance giusti, gli agenti AI possono diventare collaboratori potenti — che non dormono mai, non dimenticano mai e seguono sempre le regole che tu definisci.&lt;/p&gt;




&lt;h2&gt;
  
  
  Riferimenti
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://medium.com/p/627795029809" rel="noopener noreferrer"&gt;https://medium.com/p/627795029809&lt;/a&gt; — Architectural foundations&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://medium.com/p/da4204b3286a" rel="noopener noreferrer"&gt;https://medium.com/p/da4204b3286a&lt;/a&gt; — Constitution and guardrails&lt;/li&gt;
&lt;li&gt;OpenCode: &lt;a href="https://opencode.dev" rel="noopener noreferrer"&gt;https://opencode.dev&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;SpecKit: &lt;a href="https://speckit.dev" rel="noopener noreferrer"&gt;https://speckit.dev&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;GitHub Copilot, Amazon CodeWhisperer, Sourcegraph Cody&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Andrea Schiona — AI Breakfast / Software Architecture&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>architecture</category>
      <category>devops</category>
      <category>opencode</category>
    </item>
    <item>
      <title>AI e mondo del lavoro: rischi, numeri e nuove professioni (speciale AI Breakfast)</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 20:15:33 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/ai-e-mondo-del-lavoro-rischi-numeri-e-nuove-professioni-speciale-ai-breakfast-a12</link>
      <guid>https://dev.to/andrea_schiona/ai-e-mondo-del-lavoro-rischi-numeri-e-nuove-professioni-speciale-ai-breakfast-a12</guid>
      <description>&lt;p&gt;Il dibattito pubblico sull’intelligenza artificiale e il lavoro è polarizzato tra due narrazioni. Da un lato, l’apocalisse occupazionale: robot e algoritmi che sostituiscono milioni di persone. Dall’altro, l’ottimismo tecnologico: l’AI come forza puramente creatrice di posti nuovi. I dati delle principali organizzazioni internazionali raccontano una verità più sfumata — e più urgente. Non è una sostituzione di massa, ma una &lt;strong&gt;trasformazione disomogenea&lt;/strong&gt;: cambiano i contenuti delle professioni, emergono nuovi ruoli e si ampliano le disuguaglianze tra chi si adatta e chi resta indietro.&lt;/p&gt;

&lt;p&gt;Questo articolo analizza i numeri, i settori più esposti, le nuove professioni già visibili nel mercato e le politiche necessarie per governare la transizione, con uno sguardo specifico all’Italia e all’Europa.&lt;/p&gt;




&lt;h2&gt;
  
  
  I numeri della trasformazione: perdite e nuovi posti
&lt;/h2&gt;

&lt;h3&gt;
  
  
  WEF Future of Jobs Report 2025/2026
&lt;/h3&gt;

&lt;p&gt;Il World Economic Forum stima che entro il 2030 l’automazione e l’adozione di tecnologie AI possano rendere obsoleti &lt;strong&gt;circa 85 milioni di posti di lavoro&lt;/strong&gt; a livello globale. Lo stesso rapporto, però, ne prevede la creazione di &lt;strong&gt;circa 170 milioni di nuovi posti&lt;/strong&gt;, per un saldo positivo stimato di &lt;strong&gt;+85 milioni&lt;/strong&gt;. Il dato, però, è solo aggregato. La distribuzione geografica, settoriale e per livelli di competenza è molto diseguale.&lt;/p&gt;

&lt;h3&gt;
  
  
  McKinsey Global Institute — Agents, robots, and us (nov 2025)
&lt;/h3&gt;

&lt;p&gt;McKinsey corregge la narrazione catastrofica: non si tratta di licenziamenti di massa improvvisi, ma di una &lt;strong&gt;ristrutturazione prolungata&lt;/strong&gt; dei contenuti lavorativi. Circa il &lt;strong&gt;40% delle task&lt;/strong&gt; oggi svolte da lavoratori potrebbe essere automatizzato entro il 2030, ma solo una parte minoritaria di queste task corrisponde a intere professioni destinte a scomparire. Più frequente è il caso di ruoli che si &lt;strong&gt;ibridano&lt;/strong&gt;: un amministrativo che passa da data entry a supervisione di workflow AI, un avvocato che integra strumenti di document review automatica.&lt;/p&gt;

&lt;h3&gt;
  
  
  OECD Employment Outlook 2024
&lt;/h3&gt;

&lt;p&gt;L’OECD sottolinea che l’impatto varia drasticamente per settore e livello di qualifica. I lavoratori con &lt;strong&gt;bassa qualificazione&lt;/strong&gt; e task ripetitive sono i più esposti, mentre le professioni ad alta intensità di relazione umana, creatività e problem-solving complesso mostrano una &lt;strong&gt;resilienza strutturale&lt;/strong&gt;. L’organizzazione avverte però che senza politiche di reskilling il rischio è un aumento della disuguaglianza salariale e territoriale.&lt;/p&gt;

&lt;h3&gt;
  
  
  IMF World Economic Outlook — AI and work (apr 2025)
&lt;/h3&gt;

&lt;p&gt;Il Fondo Monetario Internazionale stima che l’AI possa contribuire a un aumento della produttività globale del &lt;strong&gt;2-3%&lt;/strong&gt; nei prossimi anni, ma con effetti distributivi asimmetrici. Nei paesi con bassa digitalizzazione — tra cui l’Italia — il ritardo nell’adozione e nella formazione amplifica il divario con le economie più avanzate.&lt;/p&gt;




&lt;h2&gt;
  
  
  I settori più esposti: chi rischia di più
&lt;/h2&gt;

&lt;p&gt;Non tutti i lavori sono ugualmente a rischio. L’automazione colpisce soprattutto le attività con queste caratteristiche:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Ripetizione strutturata&lt;/strong&gt;: inserimento dati, archiviazione, classificazione&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Regole esplicite e prevedibili&lt;/strong&gt;: procedure amministrative standardizzate&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interazione testuale semplice&lt;/strong&gt;: risposte FAQ, prima assistenza clienti, ticketing base&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I settori dove l’impatto sarà più rapido e profondo:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Amministrazione e uffici&lt;/strong&gt;: contabilità base, gestione pratiche, fatturazione, logistica documentale&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Customer service e call center&lt;/strong&gt;: chatbot e assistenza virtuale gestiscono già oggi la prima linea&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Produzione manifatturiera tradizionale&lt;/strong&gt;: robot collaborativi e AI predittiva riducono la necessità di operai su linee standardizzate&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Retail e logistica entry-level&lt;/strong&gt;: magazzinaggio, consegne, cassa automatica, inventario&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Trasporto&lt;/strong&gt;: guida autonoma per lunghe percorrenze e logistica interna&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Al contrario, risultano più resilienti le professioni che richiedono:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Intelligenza emotiva e relazione umana&lt;/strong&gt;: infermieri, insegnanti, terapisti, assistenti sociali&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Creatività e strategia&lt;/strong&gt;: designer, architetti, ricercatori, manager&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Lavoro fisico non strutturato&lt;/strong&gt;: idraulici, elettricisti, operatori sanitari&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Decisioni etiche e giuridiche complesse&lt;/strong&gt;: giudici, avvocati, revisori contabili&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  I nuovi ruoli nati con l’AI
&lt;/h2&gt;

&lt;p&gt;L’altra faccia della medaglia è la rapida comparsa di professioni che non esistevano dieci anni fa. Questi ruoli non sono marginali: richiedono competenze ibride, sono ben retribuiti e in forte crescita.&lt;/p&gt;

&lt;h3&gt;
  
  
  AI Trainer e Prompt Engineer
&lt;/h3&gt;

&lt;p&gt;Sono i professionisti che addestrano, calibrano e ottimizzano i modelli AI per compiti specifici. Non scrivono solo prompt: progettano dataset di fine-tuning, definiscono policy di comportamento, valutano la qualità delle risposte e gestiscono l’allineamento tra modello e obiettivi aziendali. In Italia la domanda è ancora inferiore all’offerta qualificata, ma sta crescendo nei settori finance, retail e pubblica amministrazione.&lt;/p&gt;

&lt;h3&gt;
  
  
  AI Governance Officer e Compliance Specialist
&lt;/h3&gt;

&lt;p&gt;Con l’&lt;strong&gt;EU AI Act&lt;/strong&gt;, entrato in vigore nel 2024, i sistemi AI ad alto rischio devono rispettare obblighi di trasparenza, documentazione e controllo umano. Questo ha creato una nuova filiera di figure professionali: il governance officer definisce le policy interne, il compliance specialist verifica la conformità legale, il risk assessor mappa i rischi algoritmici. Si stima che entro il 2027 ogni azienda con oltre 250 dipendenti avrà bisogno di almeno una figura dedicata.&lt;/p&gt;

&lt;h3&gt;
  
  
  MLOps Engineer
&lt;/h3&gt;

&lt;p&gt;Tra data science e operations, l’MLOps engineer costruisce e mantiene le pipeline che portano i modelli AI in produzione. Gestisce versionamento dei modelli, monitoraggio delle performance, aggiornamenti continui e integrazione con i sistemi aziendali. È una delle figure più ricercate nel tech, con una crescita del 70% negli ultimi due anni.&lt;/p&gt;

&lt;h3&gt;
  
  
  Data Curator e AI Content Specialist
&lt;/h3&gt;

&lt;p&gt;I modelli AI si nutrono di dati. Il data curator seleziona, pulisce, etichetta e organizza dataset di alta qualità. È una figura ibrida tra data scientist e archivista, sempre più importante in ambito healthcare, finance e media. L’AI content specialist, invece, produce e revisiona contenuti generati da AI, garantendo accuratezza, tono di marca e rispetto delle normative.&lt;/p&gt;

&lt;h3&gt;
  
  
  Red Team Specialist e AI Safety Researcher
&lt;/h3&gt;

&lt;p&gt;Quando i sistemi AI diventano critici — nella cybersecurity, nella medicina, nella mobilità — servono professionisti che li mettano alla prova in modo sistematico. Il red team specialist cerca vulnerabilità, bias, comportamenti imprevisti. L’AI safety researcher lavora invece sulla teoria e sulla pratica dell’allineamento, in contesti che vanno dai laboratori ai regolatori.&lt;/p&gt;

&lt;h3&gt;
  
  
  Figure ibride emergenti
&lt;/h3&gt;

&lt;p&gt;Oltre ai ruoli specializzati, crescono le professioni &lt;strong&gt;ibride&lt;/strong&gt;: un medico che utilizza strumenti di diagnosi assistita da AI, un avvocato che integra analisi predittiva di giurisprudenza, un giornalista che verifica contenuti generati da algoritmi, un project manager che orchestra team umani e agenzie AI. La tendenza non è la sostituzione, ma l’&lt;strong&gt;integrazione&lt;/strong&gt;.&lt;/p&gt;




&lt;h2&gt;
  
  
  Il caso Italia: tra ritardo e opportunità
&lt;/h2&gt;

&lt;p&gt;L’Italia parte da una posizione di svantaggio nella corsa all’adozione dell’AI, ma il PNRR offre risorse senza precedenti per colmare il gap.&lt;/p&gt;

&lt;h3&gt;
  
  
  Dati e analisi
&lt;/h3&gt;

&lt;p&gt;Secondo la &lt;strong&gt;Banca d’Italia&lt;/strong&gt; nel Rapporto annuale 2024, l’adozione di tecnologie digitali avanzate nel tessuto produttivo italiano è ancora limitata: meno del 20% delle imprese utilizza strumenti di AI in modo strutturato. L’&lt;strong&gt;Unioncamere Excelsior 2025&lt;/strong&gt; conferma che tra i lavoratori a rischio di trasformazione ci sono soprattutto gli impiegati del terziario e gli operai specializzati in processi routinari.&lt;/p&gt;

&lt;p&gt;Il &lt;strong&gt;CEDEFOP Skill Anticipation 2025&lt;/strong&gt; stima che in Italia tra &lt;strong&gt;3 e 5 milioni di lavoratori&lt;/strong&gt; potrebbero vedere trasformato profondamente il proprio ruolo entro il 2030. Il rischio non è la disoccupazione di massa, ma l’&lt;strong&gt;impiegificio&lt;/strong&gt;: mansioni dequalificate, salari stagnanti, working poor. Per evitarlo serve una strategia di reskilling su larga scala.&lt;/p&gt;

&lt;h3&gt;
  
  
  Il PNRR come leva
&lt;/h3&gt;

&lt;p&gt;Il Piano Nazionale di Ripresa e Resilienza destina risorse significative alla formazione digitale, alla ricerca su AI e alla transizione industriale. Il problema è la velocità di attuazione: le imprese piccole e medie faticano ad accedere ai bandi, i percorsi formativi spesso non sono allineati alle necessità del mercato. Serve un ecosistema dove università, centri di ricerca, aziende e regioni lavorino insieme.&lt;/p&gt;

&lt;h3&gt;
  
  
  Il cultural gap italiano
&lt;/h3&gt;

&lt;p&gt;Oltre alle competenze tecniche, c’è un problema di &lt;strong&gt;fiducia e comprensione&lt;/strong&gt; dell’AI. In Italia il 60% circa dei lavoratori dichiara di non conoscere gli strumenti AI, e solo il 15% li utilizza regolarmente nel proprio lavoro. Questo divario generazionale e culturale è il vero bottleneck: non si tratta solo di corsi, ma di un cambio di mentalità.&lt;/p&gt;




&lt;h2&gt;
  
  
  L’Europa e l’EU AI Act: regole che creano lavoro
&lt;/h2&gt;

&lt;p&gt;L’Europa ha scelto la strada della regolamentazione, non del laissez-faire. L’&lt;strong&gt;EU AI Act&lt;/strong&gt; classifica i sistemi AI in base al rischio e impone obblighi crescenti: trasparenza, documentazione, controllo umano, audit periodici.&lt;/p&gt;

&lt;h3&gt;
  
  
  Le nuove professioni regolamentari
&lt;/h3&gt;

&lt;p&gt;Questa regolamentazione sta generando una domanda di nuove figure professionali:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;AI Compliance Officer&lt;/strong&gt;: verifica che i sistemi rispettino le normative&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AI Auditor&lt;/strong&gt;: conduce valutazioni indipendenti su modelli ad alto rischio&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ethics Officer&lt;/strong&gt;: supervisiona l’impatto etico e sociale degli algoritmi&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Human-in-the-Loop Designer&lt;/strong&gt;: progetta interfacce e processi che mantengano il controllo umano&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Si stima che entro il 2030 in Europa nasceranno &lt;strong&gt;circa 500.000 nuovi posti&lt;/strong&gt; legati alla governance dell’AI. In Italia, se il PNRR e le politiche attive sono ben orientati, possiamo catturare una fetta significativa di questa domanda.&lt;/p&gt;

&lt;h3&gt;
  
  
  La concorrenza globale
&lt;/h3&gt;

&lt;p&gt;L’Europa non è sola nella corsa all’AI. Stati Uniti e Cina investono somme molto superiori in ricerca e adozione. Il rischio è che l’Italia e l’Europa diventino &lt;strong&gt;consumatori di tecnologie altrui&lt;/strong&gt;, senza sviluppare una filiera propria. L’alternativa è investire in ricerca applicata, startup e formazione, trasformando la regolamentazione da costo a vantaggio competitivo: la fiducia nel sistema AI europeo può diventare un asset.&lt;/p&gt;




&lt;h2&gt;
  
  
  Il vero rischio: la transizione diseguale
&lt;/h2&gt;

&lt;p&gt;L’AI non è un evento uniforme. Colpisce prima alcuni settori, poi altri; avvantaggia alcune categorie, penalizza altre; beneficia alcune regioni, lascia indietro altre. Il pericolo maggiore non è la disoccupazione totale, ma la &lt;strong&gt;polarizzazione&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Disuguaglianza territoriale
&lt;/h3&gt;

&lt;p&gt;Le regioni del Nord Italia, con una struttura produttiva più industrializzata e un tessuto di PMI più dinamico, sono meglio posizionate per adottare l’AI e per attrarre investimenti. Il Sud, con una maggiore esposizione a settori tradizionali e una minore infrastruttura digitale, rischia di subire la transizione senza beneficiarne. Senza politiche di convergenza, l’AI potrebbe ampliare il divario Nord-Sud invece di ridurlo.&lt;/p&gt;

&lt;h3&gt;
  
  
  Disuguaglianza generazionale e di genere
&lt;/h3&gt;

&lt;p&gt;I lavoratori over 50 con competenze digitali limitate sono i più vulnerabili. Le donne, sovrarappresentate in settori amministrativi e di servizio a rischio di automazione, potrebbero subire un impatto sproporzionato. Allo stesso tempo, i nuovi ruoli AI sono oggi prevalentemente maschili e giovani: serve una politica attiva di inclusione.&lt;/p&gt;

&lt;h3&gt;
  
  
  Il rischio dell’“impiegificio”
&lt;/h3&gt;

&lt;p&gt;Se la transizione non è governata, il pericolo non è la disoccupazione, ma la &lt;strong&gt;dequalificazione di massa&lt;/strong&gt;: lavoratori espulsi da ruoli tradizionali che finiscono in posizioni precarie, low-skill e low-pay. L’AI diventa allora uno strumento di compressione salariale, non di crescita. La storia ci insegna che le rivoluzioni tecnologiche creano ricchezza, ma solo se accompagnate da investimenti in capitale umano.&lt;/p&gt;




&lt;h2&gt;
  
  
  Cosa fare: strategia individuale e sistemica
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Per i lavoratori
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Formazione continua&lt;/strong&gt;: non basta un titolo di studio. Ogni 3-5 anni serve aggiornare le proprie competenze. Piattaforme come Coursera, Udacity e le iniziative PNRR offrono percorsi riconosciuti.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Competenze ibride&lt;/strong&gt;: combinare la propria expertise di dominio con conoscenze di AI. Un contabile che impara l’AI per l’audit, un marketer che impara la generazione di contenuti assistita, un insegnante che integra tool AI nella didattica.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sviluppare la “soft AI literacy”&lt;/strong&gt;: non serve essere data scientist, ma saper valutare output, riconoscere bias, gestire l’interazione uomo-macchina.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Per le aziende
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Adozione graduale e partecipata&lt;/strong&gt;: l’AI non si installa, si integra. Coinvolgere i lavoratori nella progettazione dei nuovi flussi riduce la resistenza al cambiamento e aumenta il valore della trasformazione.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Piani di reskilling interni&lt;/strong&gt;: le aziende che investono nella riqualificazione del personale hanno tassi di adozione AI superiori e minore turnover.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Governance prima della tecnologia&lt;/strong&gt;: definire policy, responsabilità e controlli prima di implementare sistemi AI ad alto rischio.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Per le politiche pubbliche
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Micro-credential riconosciuti&lt;/strong&gt;: sistemi di certificazione delle competenze AI veloci, flessibili e interoperabili tra regioni e paesi.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Lifelong learning account&lt;/strong&gt;: un fondo personale per ogni lavoratore, finanziato daaziende e Stato, da usare per formazione continua.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Politiche attive del lavoro&lt;/strong&gt;: accompagnamento alla transizione, sostegno al reddito durante la formazione, incentivi per le assunzioni in settori strategici.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ricerca e innovazione&lt;/strong&gt;: aumentare gli investimenti pubblici in AI applicata ai settori tradizionali italiani — agricoltura di precisione, turismo, manifattura, healthcare.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Il punto di equilibrio: governare, non difendersi
&lt;/h2&gt;

&lt;p&gt;L’intelligenza artificiale non distrugge il lavoro in modo uniforme. Lo trasforma. Il rischio reale è la &lt;strong&gt;transizione diseguale&lt;/strong&gt;: chi aggiorna le competenze diventa più produttivo e retribuito, chi non lo fa rischia l’esclusione.&lt;/p&gt;

&lt;p&gt;L’obiettivo non è difendersi dall’automazione, ma &lt;strong&gt;governarla&lt;/strong&gt;. Questo richiede tre cose: competenze diffuse, regole chiare e investimenti coraggiosi. L’Italia ha l’opportunità di usare l’AI per colmare i propri ritardi strutturali, ma solo se affronta la transizione come progetto collettivo, non come somma di scelte individuali.&lt;/p&gt;

&lt;p&gt;Il lavoro del futuro non è umano contro macchina. È umano &lt;strong&gt;con&lt;/strong&gt; la macchina.&lt;/p&gt;




&lt;h2&gt;
  
  
  Fonti
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;McKinsey Global Institute — Agents, robots, and us (nov 2025)&lt;/li&gt;
&lt;li&gt;WEF Future of Jobs Report 2025/2026&lt;/li&gt;
&lt;li&gt;OECD Employment Outlook 2024&lt;/li&gt;
&lt;li&gt;IMF World Economic Outlook — AI and work (apr 2025)&lt;/li&gt;
&lt;li&gt;Banca d’Italia Rapporto annuale 2024&lt;/li&gt;
&lt;li&gt;Unioncamere Excelsior 2025&lt;/li&gt;
&lt;li&gt;CEDEFOP Skill Anticipation 2025&lt;/li&gt;
&lt;li&gt;EU AI Act (Regulation EU 2024/1689)&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;🎙 &lt;strong&gt;Ascolta lo speciale podcast&lt;/strong&gt; (~5 min): &lt;a href="https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3" rel="noopener noreferrer"&gt;https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📻 Spotify: &lt;a href="https://open.spotify.com/show/033CI05HOIL6xYV16e88wV" rel="noopener noreferrer"&gt;https://open.spotify.com/show/033CI05HOIL6xYV16e88wV&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📬 Iscriviti alla newsletter: &lt;a href="https://buttondown.com/italian-ai-breakfast" rel="noopener noreferrer"&gt;https://buttondown.com/italian-ai-breakfast&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📱 Canale Telegram: &lt;a href="https://t.me/ItalianAIBreakfast" rel="noopener noreferrer"&gt;https://t.me/ItalianAIBreakfast&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Approfondimento a cura di Andrea Schiona — AI Breakfast&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>lavoro</category>
      <category>occupazione</category>
      <category>italia</category>
    </item>
    <item>
      <title>AI e mondo del lavoro: rischi e opportunità — Domenica 16/8/2026</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 20:10:35 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/ai-e-mondo-del-lavoro-rischi-e-opportunita-domenica-1682026-49cn</link>
      <guid>https://dev.to/andrea_schiona/ai-e-mondo-del-lavoro-rischi-e-opportunita-domenica-1682026-49cn</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;Approfondimento AI Breakfast — Domenica 16 August 2026&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Il grafico: posti persi vs nuovi posti
&lt;/h2&gt;

&lt;p&gt;Fonte WEF Future of Jobs Report 2025 — stime aggregate:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Posti a rischio di automazione: &lt;strong&gt;85 milioni&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Nuovi posti creati da AI: &lt;strong&gt;170 milioni&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Netto stimato entro il 2030: &lt;strong&gt;+85 milioni&lt;/strong&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Nota: il netto positivo non è automatico — dipende dalle politiche di reskilling e dalla velocità di adozione.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cosa dice la ricerca
&lt;/h2&gt;

&lt;p&gt;McKinsey Global Institute, WEF, OECD e IMF convergono su un punto: non è sostituzione di massa, è &lt;strong&gt;trasformazione disomogenea&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;In Italia, tra 3 e 5 milioni di lavoratori potrebbero vedere trasformato il proprio ruolo entro il 2030. I settori più esposti: amministrazione, customer service, produzione manifatturiera tradizionale, retail e logistica entry-level.&lt;/p&gt;

&lt;p&gt;Allo stesso tempo, nuove professioni crescono a ritmi elevati: AI trainer, governance officer, MLOps engineer, data curator, red team specialist.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cosa cambia per l'Europa e l'Italia
&lt;/h2&gt;

&lt;p&gt;L'EU AI Act introduce obblighi di trasparenza e controllo umano per i sistemi AI ad alto rischio. Questo crea domanda di nuove figure di compliance e governance. In Italia serve una strategia nazionale di reskilling: PNRR, microcredential riconosciuti e partnership tra università e aziende.&lt;/p&gt;

&lt;h2&gt;
  
  
  Il punto di equilibrio
&lt;/h2&gt;

&lt;p&gt;L'AI non distrugge il lavoro in modo uniforme. Lo trasforma. Il rischio reale è la transizione diseguale. Chi aggiorna le competenze diventa più produttivo e retribuito. Chi non lo fa rischia l'esclusione. L'obiettivo non è difendersi dall'automazione, ma &lt;strong&gt;governarla&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Fonti
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;McKinsey Global Institute — Agents, robots, and us (nov 2025)&lt;/li&gt;
&lt;li&gt;WEF Future of Jobs Report 2025/2026&lt;/li&gt;
&lt;li&gt;OECD Employment Outlook 2024&lt;/li&gt;
&lt;li&gt;IMF World Economic Outlook — AI and work (apr 2025)&lt;/li&gt;
&lt;li&gt;Banca d’Italia Rapporto annuale 2024&lt;/li&gt;
&lt;li&gt;Unioncamere Excelsior 2025&lt;/li&gt;
&lt;li&gt;CEDEFOP Skill Anticipation 2025&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;🎙 &lt;strong&gt;Ascolta lo speciale podcast&lt;/strong&gt; (~5 min): &lt;a href="https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3" rel="noopener noreferrer"&gt;https://archive.org/download/ai-breakfast/AI_Breakfast_2026-08-16.mp3&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📻 Spotify: &lt;a href="https://open.spotify.com/show/033CI05HOIL6xYV16e88wV" rel="noopener noreferrer"&gt;https://open.spotify.com/show/033CI05HOIL6xYV16e88wV&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📬 Iscriviti alla newsletter: &lt;a href="https://buttondown.com/italian-ai-breakfast" rel="noopener noreferrer"&gt;https://buttondown.com/italian-ai-breakfast&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;📱 Canale Telegram: &lt;a href="https://t.me/ItalianAIBreakfast" rel="noopener noreferrer"&gt;https://t.me/ItalianAIBreakfast&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Approfondimento a cura di Andrea Schiona — AI Breakfast&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>lavoro</category>
      <category>occupazione</category>
      <category>italia</category>
    </item>
    <item>
      <title>Test</title>
      <dc:creator>Andrea Schiona</dc:creator>
      <pubDate>Sun, 16 Aug 2026 20:03:34 +0000</pubDate>
      <link>https://dev.to/andrea_schiona/test-1dgi</link>
      <guid>https://dev.to/andrea_schiona/test-1dgi</guid>
      <description>&lt;p&gt;Test&lt;/p&gt;

</description>
    </item>
  </channel>
</rss>
