DEV Community: Paras Kavdikar

General advice on building AI agents

Paras Kavdikar — Tue, 14 Apr 2026 06:50:32 +0000

The more you constraint an AI agent, the higher is its success rate. Have multiple agents run in parallel in a cluster to reduce the scope for each agent, this doesn't guarantee the 100% success rate, but you will find it on the higher end of the spectrum.

If you are trying to build an agent for deterministic tasks, it is bound to fail sooner or later. This happens because AI can hallucinate, lose context and it is probabilistic in nature. Its a good use case if the 90-95% of success rate still helps you.

If you’re building for probabilistic tasks, you’re much more likely to get useful outcomes.

Where it works well:

Writing documentation
Summarizing Emails
Fetching important details from messy data
Drafting sales emails and outreach messages
Analyzing support tickets for patterns

Where it doesnt work well:

Final financial reporting (P&L, tax filings)
Payroll calculations
Regulatory filings
Banking transactions or ledger updates

Iterative AI agents with deterministic guardrails

Paras Kavdikar — Tue, 14 Apr 2026 06:49:58 +0000

LLMs are powerful AI models, but they can hallucinate, go out of context, or simply provide the wrong output. You need to take their output and have layers of verifications against it so that you only let the right data in. The verification layers must send the feedback for the next iteration.

The important question then becomes when to stop if it ends up in a never ending cycle? Do you stop at a certain number of retries and take the best out of all the failed outputs? or Do you keep it running until it passes a certain threshold?

It very much depends on the problem you are trying to resolve.

What are your thoughts and experiences about building iterative agentic patterns?

Ever wondered how a generic web app works in the background when you visit a website on your computer?

Paras Kavdikar — Tue, 14 Apr 2026 06:49:10 +0000

User hits a url like 'www example com' in the browser, DNS provides the browser with the ip address of the hosting server (provided by hosting services like netlify/vercel) where your frontend code lives. Then the browser does a https handshake (using tls security, enabling an encrypted tunnel) between browser and the hosting server. After the connection, the frontend/react code is downloaded and executed in the browser. It calls the node js/backend apis running on the backend servers via an http request using browser APIs like fetch/axios (The browser APIs allows javascript to interact with external sources). The backend api takes in the request, does any required database operations and then sends an http response (usually JSON) back to the react code running in the browser. The react code takes in the http response and renders the UI corresponding to the http response and you see the changes on the website.

Future of Software Engineering when AI can code. What does students and new grads need to learn?

Paras Kavdikar — Wed, 11 Jun 2025 16:08:24 +0000

Software development has become much easier since the introduction of LLMs like ChatGPT and Claude. The work that was earlier done by a whole team of developers can now be done by 2-3 mid-level engineers.
This reduces the demand for software engineers and a lot of companies are firing engineers in their teams to cut unnecessary costs. This impacts new grad software engineers the most. As most companies are cutting down costs, it is likely that fewer positions will be needed in the future.

So the question is: what should new grads learn?

The answer is not as simple, but it is a very interesting one. One thing most students and new grads have to understand is that AI and LLMs are here to stay, and we have to accept and embrace these changes rather than running away or pushing back against them. The inclusion of AI in software engineering makes it easy for developers to build, test, deploy, maintain and document applications. This allows a single software engineer to be more productive.

Some years ago, software engineers were divided into backend, frontend, DevOps, etc. But as tech kept growing, a single person started to work on both frontend and backend aspects of programming, and the new term "Full Stack Developer" was coined.

Following the same pattern, LLMs make it easier for people to become software architects. No doubt, the demand for traditional software engineers will still exist in the market because LLMs often face bugs, can't understand complex consumer requests, and cannot work with poor prompting. So engineers will still be required, but it will be nowhere close to the current supply of software engineers. This means that there will be fewer people required to know just the programming syntax.
Students should have knowledge of low-level coding, but they should now be problem solvers. They should have high-level understanding of a very wide field because AI has made it so easy to learn and understand different topics. So humans will have to do more thinking and innovation rather than diving deep into code.

Another thing newcomers can do is get good at new technologies which are not the skills of people currently in the industry. This makes older engineers less relevant in these areas, and it becomes beneficial for company's to hire more recent engineers.

I think there will be three kinds of opportunities for new students/ grads in the future:

First: Traditional Software Engineers

These will continue to exist but will be in much lower demand. These are the engineers who focus mainly on writing code, fixing bugs, and maintaining existing systems. While companies will still need them, they won't need as many because AI can handle a lot of the routine coding tasks.

Second: Creative, Broadly Knowledgeable Engineers:

This is where I see the highest demand. These engineers will have high-level understanding across multiple domains - frontend, backend, cloud, testing, embedded systems, networks, cybersecurity, data science, and more. They will be very creative and use AI tools to quickly learn new areas. Instead of just writing code, they'll focus on system design, architecture, and finding innovative solutions. They'll understand how different technologies work together and think about the bigger picture. For instance, now one developer can understand how frontend connects with backend, how cloud infrastructure supports both, how testing ensures quality, and how networks tie everything together. This way they have a better understanding of how systems work together and what are the possibilities with their combinations.

Third: Specialists in Newer Technologies:

These are engineers who specialize in cutting-edge technologies that most experienced developers haven't mastered yet. This includes things like RAG systems in AI, vector databases, prompt engineering, edge AI, blockchain systems, Web3 technologies, quantum computing basics. Since these skills are new, companies will value fresh graduates who have learned these technologies over experienced engineers who haven't.

The key is to understand that AI is changing the game. New grads who embrace AI tools, think creatively about problems, and either go broad with their knowledge or deep into new technologies will find themselves in the best position for the future of software development.