DEV Community: Bojan Tomic

I Gave My Coding Agents a Task Board: Setting Up Multica

Bojan Tomic — Wed, 26 Aug 2026 20:34:34 +0000

Running one coding agent is fine. You open a terminal, describe the job, watch it work.

Running three is where it falls apart. Each one lives in its own terminal; none of them knows what the others did, and the only record of what happened is scrollback you will close by accident. There is no answer to "what is in flight right now" other than checking four windows.

Multica is a task board for that problem. Agents get assigned issues the same way people do, they move their own cards across the board, and their work is attached to the issue instead of living in a terminal you closed.

This is what setting it up on a real project looked like: this site, its actual backlog, three agents, and what the first few hours actually consumed.

Setup: 20 minutes, no errors

I used the desktop app on macOS rather than self-hosting, which is the fastest route.

The only step that needs a decision is authentication. Log in to GitHub from your terminal over HTTPS before you start, because that is the credential the agents use when they open pull requests. Get that wrong, and everything works right up until the moment an agent tries to push, which is the worst place to discover it.

That was the whole setup. Download, sign in, connect the repository, done in about twenty minutes with nothing going wrong.

One thing that catches people out: Multica does not ship an agent. It drives the CLIs you already have. A daemon scans the machine for tools it recognizes and registers a runtime for each. It supports 23 out of the box, including Claude Code, Codex, Cursor, Copilot, OpenCode, OpenClaw, Gemini CLI, and Kimi. Mine runs on Claude Code on the MacBook. If you have no agent CLI installed, install one first.

Creating agents by asking for them

I expected a form. There is one, but it is not how you are meant to work.

Multica ships with an agent called Mika, described as a workspace chief of staff, and you create everything else by talking to it. You describe the job you want done in plain language and Mika creates the issue, the agent, or the schedule. The dialog literally prompts you with an example: "let Bojan fix the inbox loading slowness in the Web project".

Both of my other agents came out of that chat:

Agent	Job	Access
Mika	Chief of staff. Turns goals into issues, coordinates the others, builds reusable workflows	Workspace
Vera	Reviews agent-authored pull requests with fresh context	Owner only
Scout	Finds AI tools worth adding to the directory and verifies them	Workspace

The narrowness matters. An agent told to "help with the site" produces vague work. An agent told "review agent-authored PRs, and you did not write this code" produces something usable.

So does the access column. Vera is owner only, because a reviewer anyone can retask is not a reviewer.

Instructions are plain text in the agent's Instructions tab. Edit them and behavior changes on the next run; no redeploy. You are tuning a job description, not a config file.

Writing issues agents can actually finish

The board is an ordinary kanban: Backlog, Todo, In Progress, In Review, Done. Agents move their own cards. When Vera finished a review, she moved the issue to In Review herself and left the write-up as a comment.

What made this work was writing issues with an Outcome line instead of a task line. "Define the verify gate agents must pass before hand-back" is a task. "A named, runnable verify gate" is an outcome. Agents handle the second far better, because there is something concrete to check themselves against.

Some real ones off my board:

Connect the codebase to the project
Define the verify gate agents must pass before hand-back
Automated code review pass on agent PRs
Backlog sweep: turn existing TODOs and known issues into issues
Repeatable CLI triage for pending tool submissions

The part that matters: making them stop

This is the piece I would keep if I threw the rest away.

Scout runs on a schedule through Autopilot: a cron trigger at 17:00 daily, which creates an issue in a dedicated project and assigns it to Scout. The instructions are explicit about where it must stop, and that is most of the prompt:

Run one Phase 1 scouting cycle. Phase 1 only, report and stop.

Check out the repository and re-read CLAUDE.md before anything else, in particular the CRITICAL SAFETY RULES. Do not work from memory of these files.

Then STOP. Do not write SQL, open a PR, or touch the database. Set the issue status to in_review and wait.

Everything before "STOP" is read-only: sweep for candidates, deduplicate against what is already listed and anything already staged, verify each survivor by loading its own website and pricing page in a headless browser, and post the findings as a comment. Nothing is written anywhere.

I read that comment and reply naming which ones to add. That reply is what unblocks Phase 2, where it writes the migration, runs the validation scripts, and opens a pull request. Applying the migration to the live database stays my job, explicitly: "Applying it to the live database is his action, never yours."

Three stops, each waiting on a human. Silence keeps it stopped indefinitely, because there is no timeout that quietly approves for you.

That design came from a specific worry. One script in this repo writes to the production database unless you pass --dry-run, and I don't want an unattended agent editing a directory with 400+ tools.

Two smaller instructions in there earn their place too. "Never treat aggregator or roundup pricing as evidence", because that is exactly how a wrong price ends up copied across the whole internet. And "Do not pad the batch to reach a number; that is a worse outcome than a short report", because otherwise you get five results whether or not five exist.

If you take one thing from this post: decide what an agent may do without asking, and make everything else stop and wait.

What it actually consumed

Multica has an Analytics tab. This is the first few hours of use, not a month, though the dashboard window says 30 days because that is all the history there is:

	First few hours
Estimated cost	$23.43
Tasks	20, none failed
Run time	1h 15m

Read that number carefully, because it is not a bill. My runtime is Claude Code on a $ 100-a-month plan I already pay for, so nothing here appeared as an extra charge. The $23.43 is Multica's estimate of what that usage would have cost at API rates, which makes it a useful measure of how much work you got through, not money that left my account.

Split by agent, Mika did 19 tasks for an estimated $21.31 and just over an hour of run time. Vera did a single code review for $2.12 and 9.5 minutes.

The comparison worth drawing is against your plan, not against a bill. Twenty tasks in an afternoon consumed roughly a quarter of a month's allowance at API-equivalent rates. If you already pay for a coding agent subscription, that is the frame. Agents on a board burn your existing allowance faster than you typing at one of them does, and the analytics tab is where you find that out before your limits do.

Is the output any good

Here is Vera reviewing a pull request another agent wrote:

The blocking finding is the only one I'd defend without qualification, and I only have it because I ran the gate instead of reading it. No amount of staring at "eslintWarnings": 27 tells you the true count is 25; you have to execute it.

Then, unprompted:

The two I'm least sure about are the two I'd have produced from the diff alone. #1 (unenforced prohibitions) is an argument, not an observation... Both are the kind of finding a reviewer generates to look thorough, which is precisely why I'm flagging them as the suspect ones.

An agent separating what it verified from what it inferred, and telling me which half to distrust, is more useful than a longer list of findings.

I am not going to pretend every run looks like that. A backlog sweep the same afternoon canceled a dozen issues, which was correct but not exactly craftsmanship. The point is that good output is legible and bad output is visible, because it all lands on the issue rather than in a terminal.

The CLI

Everything on the board has a command-line equivalent, which matters if you want to script around it:

multica agent get <agent-id> --output json
multica autopilot get <autopilot-id> --output json
multica autopilot runs <autopilot-id> --output json

autopilot runs is the one I reach for: it is the run history for a scheduled agent, so it answers "did it fire last night and what did it do" without opening the app.

License and pricing, before you build on it

Self-hosting is free with no caps on agent count. Concurrency is limited by your hardware, and you can join multiple machines as runtimes.

Two things the marketing does not lead with.

The licence is not plain open source. It is Apache 2.0 with additional conditions: you may not offer Multica as a hosted service to third parties without a commercial license, even if you charge nothing. You may not remove the branding without a written waiver. Internal use inside one organization is explicitly fine, which covers most people reading this.

There is no published cloud pricing. The hosted version offers a free trial, and multica.ai/pricing returns a 404.

Would I keep it

Yes, with a caveat: the value is not the agents; it is the board and the gates.

The agents already existed. What changed is that I can see what they are doing, the work survives the session, and nothing irreversible happens without me saying so. That last part is what makes it usable on a live site rather than a toy repo.

Honest status: the scheduled scouting run is configured and has not fired yet, so I can tell you the design is right but not yet that it survives contact with a month of daily runs. I will report back when it has.

If you run one agent occasionally, this is a coordination layer for a problem you do not have yet. If you have three terminals open and no idea what is finished, it is the missing piece.

Originally published at intelligenttools.co, where I keep a directory of AI tools with prices checked on the vendor's own page.

I guess when you start using a harness and a model, it's difficult to change. But what would you suggest using for coding, in your experience?

Bojan Tomic — Fri, 21 Aug 2026 12:45:31 +0000

Best Namelix Alternatives for AI Logo Design in 2026

Bojan Tomic — Thu, 20 Aug 2026 13:56:43 +0000

Originally published on Intelligent Tools.

Namelix is genuinely good at the thing it was built for. You describe your idea, and it returns short, brandable coinages; unlike the keyword-mashup generators it competes with, the names actually sound like companies. It is free, filters by length, keyword, and domain extension, and learns from the names you save. If you need a name, start there.

The problem is what happens next.

Where Namelix Falls Short in 2026

The logo is a preview, not a product. Brandmark.io makes Namelix, and the polished logo beside every generated name is the top of a funnel. It looks like you are getting a name and a logo. You are getting a name and an advertisement. To download anything usable, you leave for Brandmark and pay.

Most of the good names are gone. Namelix has been running for years, similar prompts produce overlapping results, and every other founder in your category has typed roughly the same description into it. The names that feel most obviously right are the ones most likely to be taken.

Domain availability is worse than it has ever been. The generator will happily hand you a beautiful five-letter name whose .com sold for five figures in 2019. Namelix surfaces domain options, but the hit rate on an available .com for a genuinely good short name in 2026 is low enough that most people end up on a .co, .io, or .ai.

Volume beats quality by design. You get pages of names and scrolling is the interface. That works when you want to be surprised and badly when you want to decide.

None of that makes Namelix bad. It makes it a naming tool that stops one step short of what you came for. Here are the tools that cover the rest.

Looka: The Default Namelix Alternative

Looka is the tool most people land on after Namelix, and this comparison is why this post exists. Where Namelix ends at a name, Looka starts at one. You give it your company name and industry, pick styles and colors you like, and it generates full logo concepts you can edit in the browser.

What it does well: the editor is the strongest part. You can change fonts, spacing, color, layout, and icons, and the output still holds together afterward, which is not true of every tool here. The Brand Kit extends one logo into business cards, social templates, and brand guidelines.

Pricing: designing is free and unlimited. Downloading is not. The Basic package is a one-time $20 for a single PNG. Premium is a one-time $65 and is the one worth buying, because it includes vector files (SVG and EPS), multiple color variations, and full ownership. Brand Kit runs $96/year, and Brand Kit Web at $129/year adds an AI website generator.

Namelix vs Looka: not really competitors, but two halves of one job. Namelix gives you a name for free. Looka turns it into a logo you own for $65. The catch is the free-to-design, pay-to-download model: you invest an hour in the editor before finding out you care enough to pay. That is deliberate, and it works.

Brandmark: Namelix's Own Paid Half

Brandmark is where Namelix sends you, so it is worth evaluating on its merits rather than treating it as an upsell.

What it does well: output quality is a step above the template-driven tools, and the pricing is the most honest here. You pay once and keep access, with full copyright and unlimited revisions. No subscription quietly renewing next year.

Pricing: Basic is $35 one-time for PNG and vector files. Designer is $95 one-time and adds source files, brand guidelines, a one-page website, business card, and social designs. Enterprise is $195 one-time and includes up to 10 hand-crafted concepts plus font downloads.

Namelix vs Brandmark: this is the intended path and a reasonable one. If you already like the preview Namelix showed you, Brandmark is the shortest route to owning it. It is not my first recommendation on price-to-flexibility: $95 for the tier most people need beats Looka's $65 Premium, and Looka's editor gives you more room afterward.

Canva: Best AI Logo Maker If You Already Design

Canva is not a logo generator. It is a design platform with a logo maker, and that distinction decides whether it is right for you.

What it does well: everything after the logo. Your logo lives in the same workspace as your pitch deck, social posts, invoices, and thumbnails, with brand colors and fonts applied consistently. No other tool here touches that. AI generation is competent rather than remarkable, but the template library is enormous.

Pricing: the free tier is genuinely usable and lets you make and download a logo. Pro is roughly $15 to $18 per month depending on region and billing period, and Business runs about $25 per user per month. Check the current number before you buy, because Canva's regional pricing moves.

Namelix vs Canva: Canva will not name your company, and its logo output is more assembled than designed. What it gives you instead is the other forty things you need in month one. For a solo founder already making their own social graphics, the Pro subscription does far more work than a one-time logo purchase.

One caveat worth knowing: logos built from Canva's stock elements carry usage restrictions and generally cannot be trademarked. If you plan to register the mark, read the license terms first or use a tool that grants full copyright.

Tailor Brands: When the Logo Is Not Really the Point

Tailor Brands has quietly stopped being a logo company. It is now an LLC formation service that includes a logo maker, and its pricing reflects that.

What it does well: it handles company registration, operating agreements, annual compliance, a domain, a website, and the logo in one flow. At the "I need to actually exist as a business" stage, that bundle removes real work.

Pricing: Lite is $0 plus state fees and includes 8 logos, a domain, a website, and business card tools. Essential is $199/year plus state fees and adds accelerated processing and legal compliance. Elite is $249/year plus state fees.

Namelix vs Tailor Brands: barely comparable. Namelix is a free naming tool; Tailor Brands is a business formation service with branding attached. Judged purely as an AI logo generator, it is the weakest here, and the logos are visibly template-driven. Judged as a way to go from idea to registered company in an afternoon, the logo is a bonus.

Kittl: For People Who Can Actually Draw

Kittl is the outlier here, and the one I would point a designer toward.

What it does well: it is a real design tool with serious typography, vintage and hand-lettered styles, and effects that the generator-driven tools cannot produce. Logos out of Kittl look made rather than assembled. The AI features are token-metered rather than unlimited.

Pricing: a free tier gives 200 one-time AI tokens. Pro is $15/month, or $12/month billed annually, with 2,000 AI tokens per month per editor. Expert is $35/month ($26 annually) with 12,000 tokens, and Max is $65/month ($48 annually).

Namelix vs Kittl: opposite philosophies. Namelix removes every decision; Kittl hands them all back. If you have design skills and want output that looks like a person made it, this is the best tool here. If you want a logo in ten minutes with no opinions about kerning, it is the wrong one.

Logo Diffusion: The Actual AI Logo Generator

Logo Diffusion is the only tool here that does image-model generation rather than template assembly, which makes it the most interesting and least predictable.

What it does well: you can generate from a text prompt, a rough sketch, or an existing image, a workflow none of the others offer. Sketch-to-logo genuinely works. It includes a vectorizer, background remover, upscaling, and editing models.

Pricing: a free tier gives 24 credits, roughly 12 logos, with no commercial rights or vector export. Basic is $24/month for 1,000 credits. Pro is $49/month for 2,500 credits and adds concurrent generation and upscales. Elite is $99/month. There is also a $15 one-time Starter Pack with 300 credits and three exports, which is the sensible way to trial it. Annual billing takes 20% off.

Namelix vs Logo Diffusion: the closest thing to Namelix in spirit; both generate volume and let you fish. Expect the same trade-off, with some genuinely original results, plenty of unusable ones, and the occasional mangled letterform image models still produce. Subscription pricing is the drawback for a one-off logo, which is what the Starter Pack solves.

Design.com: Fast, With a Pricing Asterisk

Design.com is an AI logo maker wrapped in a wider design platform covering business cards, letterheads, social templates, and a website builder.

What it does well: speed. Name in, usable concepts out, minimal decisions. The asset range is respectable for its tier.

Pricing: this is the caveat. Design.com does not publish a pricing page. It advertises free logos and free business cards, but it doesn't disclose the cost of downloading production files until you are inside the flow. Every other tool here tells you the price up front.

Namelix vs Design.com: similar generate-and-browse experience with more finished output. I would reach for Looka or Brandmark first, purely because I know what they cost upfront.

Two More Namelix Alternatives

Shopify Business Name Generator is free and competes with Namelix directly rather than with the logo tools. Output is more literal and less brandable, but it checks domain availability inline, the exact friction Namelix leaves you with.

Hatchful, also from Shopify, is a free logo maker producing social and branding asset packs. Quality is template-obvious, and customization is limited, but for a side project you are testing for a month, free is the correct price.

Comparison Table

Tool	Best for	Pricing	Own the files?
Namelix	Naming only	Free	No logo files
Looka	Most people	$20 or $65 one-time, $96/yr kit	Yes, on Premium
Brandmark	Buy once, keep forever	$35 / $95 / $195 one-time	Yes, full copyright
Canva	Founders doing their own design	Free, Pro approx $15-18/mo	Restricted license
Tailor Brands	Forming the company too	$0-$249/yr plus state fees	Yes, on paid tiers
Kittl	Designers	Free, $12-$15/mo Pro	Yes
Logo Diffusion	Sketch-to-logo, novelty	$15 one-time or $24-$99/mo	Paid tiers only
Design.com	Speed	Not published	Unclear until checkout

Which One Should You Actually Use

If you need a name: stay on Namelix, and check availability on Shopify's generator or a registrar before you fall in love with anything.

If you need one logo and want to stop thinking about it: Looka Premium at $65. Vector files, color variations, full ownership, one payment. The default answer for most people here.

If you hate subscriptions: Brandmark Designer at $95 one-time. Pricier than Looka, but you own it, and there is nothing to cancel.

If you will be making your own marketing anyway: Canva Pro, treating the logo as one output of a tool you were paying for regardless. Check the trademark limitations first.

If you can design: Kittl. Nothing else here will produce work that looks handmade.

If you are registering a company this week: Tailor Brands, accepting that the logo is the least impressive part of a bundle doing more valuable work elsewhere.

The uncomfortable summary is that the best AI logo maker in 2026 still only gets you to 80% of a competent mark, and the last 20% is either a designer or your own taste. What has improved is the price of that 80%. A one-time $65 for vector files you own is a reasonable trade for a business with no brand budget yet.

The AI Bubble Looks Exactly Like January 2000

Bojan Tomic — Thu, 19 Feb 2026 10:56:07 +0000

Super Bowl LX just happened. Nearly a quarter of the ads were for AI companies. OpenAI, Anthropic, Google, Amazon, Meta—all spending $8-10 million per 30-second spot.

Companies that burn billions in losses pay millions to tell you their products are revolutionary.

I've seen this before. So have you, if you were coding in 2000.

The Dot-Com Parallel Nobody's Talking About

January 2000: A leaked essay called "bubble.com" warned that internet companies were catastrophically overvalued. Nobody listened. Two months later, the market peaked and then lost 75% of its value over 18 months.

Super Bowl XXXIV (January 30, 2000) was packed with dot-com ads. Pets.com with the sock puppet. E*Trade. Seventeen different internet startups are burning venture capital on advertising.

Pets.com had its IPO eleven days after the Super Bowl. The company shut down nine months later.

The Numbers Don't Make Sense

OpenAI is reportedly discussing a funding round that would value it at $830 billion—more than Coca-Cola, Chevron, or Costco.

The company projects:

$14 billion in losses for 2026
$115 billion in cumulative losses through 2029
$20 billion in revenue against $1.4 trillion in compute commitments

That's 1.4% of commitments covered by revenue.

HSBC analysis: Even if OpenAI hits $200 billion in revenue by 2030, they'll still need an additional $207 billion in funding to stay in business.

Microsoft lost $440 billion in market capitalization in a single day after investors questioned its AI spending plans.

The Debt Bubble Under the AI Bubble

Companies supplying infrastructure to OpenAI have taken on $96 billion in debt to fund operations. The big five tech companies (Amazon, Google, Meta, Microsoft, Oracle) added $121 billion in new debt in 2025 alone—four times their historical average.

Oracle is being sued by bondholders for concealing the amount of debt they'd need for an AI buildout. Their credit rating is approaching junk status.

This is exactly how bubbles work: companies with unproven business models borrow unsustainable amounts, then spend desperately on advertising to create the appearance of legitimacy.

The Tools Aren't Going Away

I use Claude Code daily. AI coding assistants are genuinely useful when used correctly. But they're tools that require human judgment, not replacements for it.

The valuable tools will survive the crash. The rest won't. Companies that burn billions while losing money won't make it. The ones with sustainable unit economics might.

What This Means for Developers

Short term: Keep using AI tools while they're heavily subsidized. Learn what works and what doesn't. Build things you couldn't build before because the tedium was too high.

Long term: The opportunities after the crash are probably more interesting than what we're working on now. The dot-com crash gave us Google, Amazon, and modern web development. The AI crash will leave us with whatever actually works.

The code still needs to be written. The bugs still need to be found. The architecture still needs human judgment.

That's not changing, regardless of how many AI companies burn through their funding. That said, I am writing this article in a blog pair-programmed with Claude. Grammarly checked it and offered suggestions. AI is everywhere, and it can't be stopped, no matter the economy.

And yeah, the image is also AI 😄

Originally published at intelligenttools.co

I Let Claude Code Run Unsupervised for 8 Hours - Here's What It Built

Bojan Tomic — Fri, 16 Jan 2026 22:19:27 +0000

Last week, I discovered the Ralph Wiggum technique for Claude Code. Named after the perpetually confused Simpsons character, it lets Claude iterate on code autonomously until tests pass.

I gave it a real production task: refactor our healthcare app's authentication module (8 files, 1,200 lines, JWT handling, session management, password reset).

Started: 11 PM Friday

Woke up: 7 AM Saturday

Result: 47 commits, auth module fully refactored, test coverage improved from 62% to 87%

Cost: $23.14 in API credits

Time saved: 6-8 hours of my weekend

The Setup

claude
> /plugin install ralph-loop@claude-plugins-official
> /ralph-loop "Refactor auth module. Extract business logic from controllers. 
   Max 20 lines per function. Coverage 80%+. All tests must pass."
   --max-iterations 100

Then I went to sleep.

What Actually Happened

Claude ran 47 iterations, each one:

Making changes
Running tests
Reading failures
Adjusting approach
Trying again

When tests finally passed at iteration 47, it stopped with <promise>COMPLETE</promise>.

The Real Results

Worked:

Business logic cleanly separated from controllers
All 8 files refactored systematically
Test coverage jumped to 87%
Zero TypeScript errors
Clean, consistent commit messages

Didn't work:

Still had to review everything (found 2 edge cases it missed)
Cost analysis wasn't perfect ($23 vs my estimate of $15)
Some function names were... creative

The Prompt Makes Everything

My first attempt with a vague prompt ran for 30 iterations and produced garbage.

What worked:

Clear success criteria (tests pass, coverage %, linting)
Explicit constraints (max line length, style rules)
Step-by-step process
Definitive exit condition

What didn't:

"Make it better"
"Improve code quality"
Subjective goals with no metrics

When to Use This (and When Not To)

Perfect for:

Refactoring with good test coverage
Adding tests to untested code
Bug fixes with failing tests
Code cleanup tasks

Don't use for:

Anything without tests (Ralph has no feedback)
Security-critical code (needs human review at each step)
Subjective work (UI design, naming, documentation)
Production data (run in isolated dev environments only)

The Bottom Line

This isn't AGI. It's an extremely capable pattern matcher with good self-correction when tests provide clear feedback.

For healthcare software with HIPAA requirements, I'm keeping autonomous loops in dev environments only. But for the right tasks? It's like having a junior developer who works 24/7 for $20/day.

Full writeup with detailed breakdown, failures, lessons learned, and exact prompts:

👉 Read the complete article on intelligenttools.co

Questions:

Have you tried autonomous coding loops?
What's your experience with Claude Code or similar tools?
What guardrails would you add to this approach?

Drop your thoughts below - genuinely curious how others approach this.

AI Coding in 2026: 10 Predictions

Bojan Tomic — Fri, 26 Dec 2025 11:38:46 +0000

I spent 18 months testing AI coding tools in production and analyzed Stack Overflow's 2025 AI Developer Survey (71,000+ developers, 76% now using AI for code generation).

After testing Claude Code, Cursor, GitHub Copilot, and 17 other tools on genuine healthcare software, here are my predictions for what actually happens in 2026.

Some of these will make you uncomfortable. Good.

🚨 Junior devs stop writing boilerplate entirely

By late 2026, companies will test for "AI collaboration skills" instead of syntax memorization. Interview questions shift from "Write a function to sort an array" to "How would you prompt Claude to build this feature?"

The more established your tech stack, the faster this hits. I'm already seeing it with Node.js/TypeScript.

💰 The "AI-First Developer" becomes a job category

New role: 80% directing AI agents, 20% writing critical logic. Salary range: $120k-180k. One AI-first developer replaces 3-4 traditional juniors on greenfield projects.

This isn't about replacing developers. It's about creating force multipliers who know how to orchestrate AI tools while maintaining code quality.

🔐 First major security breach from AI-generated code

A high-profile incident stems from AI-generated code with a subtle vulnerability. Trust in AI tools drops from 44% to sub-30% for 2-3 months before recovering with better tooling.

This is inevitable. AI tools generate millions of lines of code daily. Eventually, one will cause a significant incident.

💸 Pricing flips from subscription to usage-based

Cursor's $20/month unlimited becomes unsustainable as compute costs surge. By mid-2026, most tools switch to $0.02-0.05 per request. Power users generating entire backends pay $300+/month.

The economics don't work for flat-rate unlimited access when some users generate 100x more tokens than others.

Read All 10 Predictions

I've shared 4 of 10 predictions here. The complete analysis includes:

Why Claude Code captures 30% of the CLI-based market
MCP Protocol becoming the connector standard (5,000+ servers by Q4)
Tool fragmentation by developer specialty (frontend vs backend vs DevOps)
Open-source AI tools reaching feature parity
First AI coding tool IPO (Cursor at $5B+ or Copilot spins out)
Prompt engineering is becoming a 2-3 hour/week skill

Each prediction includes my reasoning, timeline, and what developers should do to prepare.

👉 Read the full article with all 10 predictions and analysis →

Methodology

Based on:

Stack Overflow 2025 AI Developer Survey (71,000+ devs)
18 months testing 20+ AI coding tools
100+ conversations with developers in production
My own production use at ICANotes (healthcare software)

Not based on: YouTube demos, marketing materials, or Twitter hype.

About Me

I'm Boki, a senior full-stack developer in Serbia working on healthcare systems. I run intelligenttools.co where I share honest, technical AI tool reviews from a developer's perspective.

What's your boldest prediction for AI coding in 2026? Drop it below 👇

--- title: 10 Bold Predictions for AI Coding Tools in 2026 published: true description: 10 predictions from 18 months of production testing (some will make you uncomfortable) tags: ai, webdev, predictions, coding cover_image:

Bojan Tomic — Fri, 26 Dec 2025 11:34:17 +0000

ChatGPT vs Claude Code: Over a Year of Building Production Software

Bojan Tomic — Sat, 20 Dec 2025 12:04:21 +0000

Quick intro

I've spent over a year using both ChatGPT and Claude Code CLI daily while building healthcare software. Not for side projects or demos - for actual production systems where bugs matter.

This isn't a benchmark comparison. Those tell you nothing about real work. This is what it's actually like to use both tools when you're shipping code that matters.

Spoiler: They're not competing. They solve different problems.

ChatGPT vs Claude Code: Over a Year of Building Production Software

I've been using AI coding tools for about a year now, and for the last 6 months, Claude Code CLI has been running in my terminal every single day. But I'm not here to tell you it's perfect or that you should switch. I'm here to tell you what actually matters when you're building real software.

Because here's the thing: ChatGPT and Claude Code aren't really competing. They're solving different problems.

The Setup

I'm a senior full-stack developer at a healthcare software company. I build clinical systems with Node.js, MySQL, and AWS. My typical day involves fixing bugs in production, adding features to existing systems, and reviewing code from our team.

For the past year, I've used Claude Code CLI as my primary coding assistant. I also use ChatGPT (with the new GPT-5 models) for specific tasks. This isn't a benchmark comparison – those tell you nothing about real work. This is what it's actually like to use both tools when you're shipping code that matters.

What They Actually Are

ChatGPT is a web interface for chatting with an AI. You can describe problems, paste code, and get suggestions. With the new GPT-5, GPT-5.1, and GPT-5.2 models, it's legitimately smart at coding. OpenAI also has Codex (their cloud coding agent), but I'm comparing ChatGPT's web chat interface here.

Claude Code CLI is a terminal tool. You run it in your project directory, and it can see your entire codebase, modify files, run git commands, execute bash scripts, and use your CLI tools. It's powered by Claude Sonnet 4.5 (I can switch to Opus 4.5 for complex stuff). It's not a chat interface – it's integrated into your actual development workflow.

That difference is everything.

[Continue with the rest of my blog post content here: https://intelligenttools.co/blog/chatgpt-vs-claude-code-1-year ]