Every AI Has a Personality

Dear Reader,

One afternoon I asked the same hard question to seven different AI systems — not a toy prompt, a real product problem for kunwersachdev.com.

I expected seven variations of the same answer. What I got felt like seven people walking into a room:

One wanted to rewrite the brief before writing a line of code. One opened the codebase and started scaffolding. One brainstormed three strategies and asked which market I was really serving. One returned sources with citations before offering an opinion. One pulled context from the documents already living in my Google world. One shipped a pull-request-shaped answer that assumed I lived inside Microsoft and GitHub. One answered with the blunt urgency of a late-night timeline — sharp, current, occasionally reckless.

That was the day I stopped asking “Which AI is best?” and started asking “Which AI belongs on this job?” — the same shift I described in The Day AI Became My Translator.

Why they feel different

People talk as if every large model is climbing the same mountain with the same boots. They are not.

Some labs optimize for conversation and general usefulness. Some optimize for tool use inside an editor. Some optimize for search, freshness, and receipts. Some optimize for the product ecosystems where work already lives — Google Workspace, GitHub, Microsoft 365, or the live stream of X. The destination can look similar: “help the human.” The training choices, safety knobs, memory, tool wiring, and product incentives do not.

Same goal. Different algorithms. Different personalities. That is not a bug — it is the market.

A founder who treats AI like one search box will keep getting disappointed. A founder who leads AI like a team — the lesson in Why AI Still Needs Human Judgment — starts winning.

Seven characters who walked into my room

ChatGPT — the versatile generalist

Strength: range. Weakness: it can sound finished before the thinking is finished.

ChatGPT is the colleague who can join almost any meeting. Draft, debate, outline, role-play a customer, stress-test a pitch — it will show up. Its place is breadth: the first whiteboard, the messy middle, the “talk this through with me” hour. The risk is fluency. Confidence is not the same as correctness. I use it to open doors, not to lock decisions.

Claude — the careful thinker

Strength: judgment and long-form care. Weakness: it may slow a sprint that only needed speed.

With Claude, I often feel less like I am prompting a machine and more like I am speaking with someone who wants the problem defined before the solution is sold. Writing, strategy, product framing, rewriting an essay until the spine is honest — that is its chair at the table. When I need a careful second brain, I start here. When I only need a quick stub, I do not force it to hold a board meeting.

Perplexity — the research scout

Strength: search with citations. Weakness: it is a scout, not your strategist-in-chief.

I do not ask Perplexity to dream my product for me. I ask it to verify what I think I already know. In business, confidence without verification is expensive. Its place is the briefing room before the boardroom — sources first, opinion second.

Gemini — the ecosystem native

Strength: lives where Google work already lives. Weakness: outside that gravity well, another tool may lead.

When the work sits in Docs, Sheets, Gmail, or Workspace, Gemini often wins on proximity. Multimodal reach and Google-stack context matter more than abstract “IQ contests.” Sometimes the best AI is not the one that sounds smartest in a demo. It is the one already standing in the room where your files are.

Cursor — the IDE craftsman

Strength: builds inside the codebase. Weakness: without a clear brief, it will build the wrong beautiful thing.

Cursor did not magically make me a programmer. It gave me the confidence to participate in building. When I explain an idea clearly, it turns intention into files, diffs, and momentum. When it drifts, I interrupt: “Stop. Why this way?” Challenged, it often finds a cleaner path. Its place is execution — the gap I wrote about in The Ideas That Never Got Built. Start with The Question Before the First Line of Code.

Copilot — the ship-inside-the-workflow partner

Strength: Microsoft and GitHub gravity. Weakness: it inherits the habits of the workflow you already have.

GitHub Copilot — and the wider Microsoft Copilot family — is built for people who already live in repos, PRs, VS Code, and Office. Its personality is practical: autocomplete the next honest step, draft the PR description, keep the conveyer moving. Its place is institutional velocity. The weakness is subtle: if your process is noisy, Copilot will cheerfully amplify the noise. Bad process plus a fast assistant is still bad process.

Grok — the blunt realtime voice

Strength: edge and X-native freshness. Weakness: blunt is not the same as wise.

Grok walks in without polishing every sentence for a conference room. Wired into the live pulse of X, it can feel current when other systems still sound like last quarter’s briefing. I use that energy when I need an unvarnished take or a sense of what the timeline is arguing about today. I do not hand it the final board memo without a second pass from a calmer mind on the roster.

Same destination, different routes

Here is the story underneath the brands: OpenAI, Anthropic, Google, Perplexity, Cursor, Microsoft/GitHub, and xAI are not failing to copy each other. They are placing different bets — on chat, on judgment, on search, on editors, on enterprise workflow, on realtime culture.

So of course the personalities diverge. If you optimize a model for tool-calling inside a repository, it will not sound like a citation scout. If you optimize for Workspace context, it will not feel like a midnight X argument. Different training data, different reinforcement, different product surfaces — different colleagues.

One afternoon I asked Claude to improve an article, ChatGPT to challenge it, Perplexity to verify facts, Cursor and Copilot to implement, Gemini to tidy the Workspace trail, and Grok to tell me what the live conversation sounded like. Suddenly I was not “using AI.” I was conducting a team.

The lesson

People ask, “Which AI should I subscribe to?” The better question is: which AI should carry this responsibility today? Great entrepreneurs do not hire one person to run a company. They build teams. The future of AI looks the same — and execution still beats knowledge alone, as I argued in Why AI Is Taking Jobs.

Don’t marry one model. Assemble the roster. Every AI has a personality — and every one has a place.