Your AI is a brain in a jar

A client showed me his "AI strategy" last month. It was three people on his team pasting things into ChatGPT and copying the answers back out. Quotes, emails, compliance summaries — all manually ferried between a chat window and the actual systems where the work lives.

He wasn't doing it wrong. That's genuinely what ChatGPT is: you bring it information, it thinks, it gives you back words. It can write a contract summary, draft a client email, explain a regulatory requirement in plain English. Impressive — genuinely.

But here's what it can't do: anything.

It can't open your inbox. It can't pull last month's figures from Xero. It can't update a client record or file a document. It can think about all of those things. It just can't touch them.

ChatGPT is a brain in a jar. Extremely capable, completely isolated. A useful thing to have in a drawer — but not what people mean when they talk about AI changing how a business operates.

For that, the brain needs a body.

What a body looks like

Think about what makes a new hire effective. It's not just what they know — it's what they can access, what they can operate, and how they've been trained to handle your specific situations.

• Eyes — they can read your documents, scan data, check an email
• Hands — they can update a record, send a message, move a file, write a report
• Ears — they notice when something happens: a new enquiry arrives, an invoice goes overdue, a threshold gets crossed
• Memory — not textbook knowledge, but memory of your business. Your clients, your processes, what happened last time
• Training — they've been taught how you do things. Your way of handling a quote, your tone in client emails, your criteria for prioritising jobs
• Rules — they know what they're allowed to do, in what order, and when to stop and ask a human

The brain is the commodity — everyone has access to the same models. The body is what makes it yours.

The labs are telling you this, quietly

Look at what every major AI lab has shipped in the last twelve months. Anthropic built Claude Code — a harness for developers. OpenAI built Cowork and Projects — harnesses for knowledge work. Google shipped Gemini CLI. They've all seen the same thing from the inside: a chat window isn't enough, and the next tier of value lives in what wraps the model.

That's the piece most business owners are still missing.

When you open ChatGPT, you're not using "AI" in some abstract sense. You're using OpenAI's chosen presentation of the underlying model — their personality, their guardrails, their pre-decided context limits. Claude is the same story. Different personality, different defaults — same underlying constraint.

For the model to do anything operational, something has to wrap around it.

Chat → Integrated → Agent

A useful way to see the spectrum.

At one end: the chat window. claude.ai, chatgpt.com. Someone else's body, broadly tuned. Useful, but generic.

In the middle: integrated. Claude Code, ChatGPT Cowork, Claude Projects. Still the provider's body, but with project-level persistence, custom tools, and richer context. This is where most teams graduate to when they outgrow chat. The providers built these precisely because they could see people hitting the ceiling.

At the far end: agents. You own the body. Persistent memory of your operation, trained on your processes, wired into your systems, permissioned to your rules. The model inside is a commodity. The body is the product.

The distance between chat and agent isn't how smart the model is. It's how much of the surrounding system you control.

What this looks like in practice

Say you run an engineering consultancy. Enquiries come in by email. Right now, someone reads them, logs them in a spreadsheet, drafts a response, and maybe updates the CRM if they remember.

An agent doing that job has eyes on the inbox. When an enquiry arrives, it reads the email, classifies it against your service types, drafts a response in your firm's tone, logs the details in your CRM, and flags anything unusual for a human to review. It does this at 2am on a Sunday if that's when the email lands.

That's not a smarter brain. It's the same brain with a body built for a specific job.

The important bit: each capability — reading email, writing responses, updating the CRM, knowing your tone — is a separate connection. You don't build the whole thing on day one. You start with one capability, prove it works, then add the next. An agent is modular in the same way a team is modular: you add capacity where you need it, when you've earned confidence that the last addition is pulling its weight.

The questions worth asking

If the question in your head is which AI should we use?, you're asking the wrong question. It's like asking whose brand of pipe cutter you should buy when the real question is whether you have plans for the house.

The right questions are body questions:

• What does the system need to see?
• What should it be able to do on its own?
• What events should trigger it?
• What does it need to remember across time?
• What does it need to know about your specific operation?
• What may it do alone, and what needs sign-off?

Answer those six and the model choice answers itself. Probably whichever one you can access cheapest today.

A good time to be a tinkerer

Two years ago you'd pay an agency six months and a six-figure budget to build any of this. Today, a small team with API keys and the right discipline can ship something genuinely useful in weeks.

That's not because the models got smarter (though they did). It's because the tinkerer's tools finally caught up. Frameworks. Evals. Skill libraries. Agent runtimes. The cost of building a body has collapsed.

If you're running a business with 10–40 people and you've been watching the AI headlines feeling vaguely behind, here's the thing: you don't need to pick the right model. You need to build the right body.

Your business is the body. Everything else is commodity.

The question worth sitting with isn't "should we use AI agents?" It's the older, more useful one: which parts of your operation would benefit from a trained person who can see your data, act on your systems, and work without being asked — but who always follows the rules you've set?

That question has been worth asking since before AI existed. The technology just changed the economics of the answer.

Karl Howard · Reforged · 10 March 2026