Just Talk To It

Published: 2026-06-03 · 5 min read

Peter Steinberger wrote a piece called Just Talk To It.

He is mostly talking about software development. He builds with these tools all day, so naturally the article is written for programmers, agent builders, and people already deep in the weeds.

But I think the bigger point is not really about coding.

The tools are now good enough that normal people can create with them.

That is the part people should be paying attention to.

A lot of the conversation around AI coding agents is framed as doom. Software developers are going to be replaced. The economy is going to be disrupted. The old jobs are going away. There is truth in some of that. If your job is mostly rote work, and the tools can now do the rote work, you should take that seriously.

But I think the better analogy is closer to the printing press.

When books became easier to produce, the people whose job was manually copying books were disrupted. That was real. But the larger story was not simply that one class of labor changed. The larger story was that reading and writing spread. More people could access ideas. More people could participate. Literacy expanded.

I think something similar is happening with software and creation now.

The skill is moving.

It is not disappearing. It is moving away from “can you personally write every line of code?” and toward “can you explain the problem clearly enough, give feedback, make judgment calls, and push the tool toward something useful?”

That is a massive shift.

It means the contractor with a real job-site coordination problem can start building toward a solution. It means the business owner who has lived inside a broken process for ten years can finally start turning that experience into software, systems, workflows, or agents.

Not because he suddenly became a traditional programmer.

Because the tools changed what is possible.

The blue-collar creator era

Think about a contractor on a job site.

He is coordinating with other contractors, crews, materials, customers, and schedules. He knows exactly where the work breaks down because he has lived it for years.

Maybe everyone is waiting on one trade to finish before the next person can come in. Maybe nobody knows whether the job is actually ready. Maybe the customer gets a different answer depending on who they ask. Maybe the owner has said a hundred times, “It would be nice if there was something that kept everybody on the same page.”

Historically, that person probably was not going to build the tool.

He understood the problem better than almost anyone, but he did not have the technical skill set to turn it into software.

That gap is shrinking.

Now he can open Claude or ChatGPT or Codex and start talking through the problem. He can explain the workflow. He can describe what each person needs to know. He can sketch the texts, the statuses, the handoffs, the alerts, the weird exceptions that only someone in the field would know.

And the model can help him move toward something real.

Maybe it starts as a checklist. Maybe it becomes a spreadsheet. Maybe it becomes a simple app. Maybe it turns into a set of agents that keep track of the work and send the right updates at the right time.

The point is not that every contractor is suddenly going to become a software company.

The point is that people with deep, practical knowledge can now create things they could not create before.

That is why “just talk to it” matters.

The first move is not becoming technical.

The first move is explaining the work.

The model needs more than a shallow prompt

This is also why a lot of people try AI once and leave disappointed.

They give it almost nothing to work with.

“Write me an email.”

“Make this better.”

“Help me automate my business.”

Then they get back something that sounds like it came from the internet blender and assume the whole thing is overhyped.

Sometimes it is overhyped.

But a lot of the time, the problem is that the model was given a shallow version of the work.

The useful version sounds more like this:

“I run a contracting business. Here is how jobs move from estimate to scheduling to completion. Here is where customers get confused. Here is where my crews usually miss the handoff. Here are the texts I send when I am trying to clean it up. Help me figure out what part of this should become a repeatable process.”

That is a different conversation.

Now the model has context. It has constraints. It has examples. It has the actual shape of the problem.

These tools are not at their best when you treat them like a magic form field. They are at their best when you talk to them like a thinking partner and keep pushing.

Explain the issue. Correct the answer. Add more context. Ask it what is missing. Ask it what would break. Ask it what should stay human.

That back-and-forth is where the value starts to show up.

Context compounds

The next layer is where this gets even more interesting.

A blank chat can be useful. But an agent that knows your business is much more useful.

If the tool knows your website, your files, your offers, your customers, your voice, your mission, your past decisions, and what you are trying to build, the quality of work changes.

You do not have to re-explain everything every time.

The context starts to compound.

That is the difference between asking a random assistant for a generic answer and working with a system that has been learning how you operate.

This is a big part of what I have built for myself and what I help clients move toward. At first, you may just be learning how to talk to the tools. Then you start giving them better context. Then you start building repeatable processes. Eventually, some of the work can move through agents that know enough about the business to do useful work without starting from zero every time.

That does not happen by accident.

You have to decide what the agent can see. You have to decide what it should remember. You have to train the voice. You have to define the job. You have to verify the output. You have to keep the system from drifting into slop.

But once it starts working, the leverage is real.

You can give a small nudge and get a lot more movement than you used to.

This is not just for software people

An online store owner has a hundred little steps in the work.

Listing products. Writing descriptions. Sending emails. Creating invoices. Handling order updates. Reporting on what sold. Testing new promotions. Answering customer questions.

If the owner knows the job, those steps can be taught.

Not all at once. Not perfectly on day one. But piece by piece.

The same is true in a service business. Lead intake, follow-up, quote review, scheduling, customer updates, internal reporting — these are jobs with patterns. If you understand the pattern, you can start teaching the tools how to help.

That is the practical version of agents.

Not some sci-fi workforce replacing everyone overnight.

A set of workers with narrow jobs, enough context, and a clear standard for what good work looks like.

You may talk to one main agent, but underneath that you can have other agents handling specific pieces of the operation. One watches for leads. One drafts follow-ups. One summarizes calls. One prepares reports. One checks whether something fell through.

The business owner should not have to manage all of that complexity alone.

That is part of where Bot Doctor fits.

Where I can help

There is a lot of noise right now.

Every tool says it will change your business. Every platform is adding AI. Everybody has a demo.

Most business owners do not need more noise. They need someone to help them figure out where the leverage actually is.

That might start with coaching and education: how to talk to the models, how to prompt better, how to give useful context, how to know when the answer is bad.

It might be a diagnostic: where in the business would these tools actually help first?

It might become a managed agent setup: a main agent you talk to, with worker agents underneath it doing specific jobs and staying maintained over time.

The shape depends on the business.

But the starting point is the same.

Talk to the tools.

Bring them a real problem.

See what they can do.

Do not be afraid to experiment right now. This is the time to build, try things, learn faster, and figure out where these tools fit into your work.

And if you run a business and feel like there is something here for you, but you do not know where to start or how to make it useful, that is what I am here for.

I can help you learn how to talk to the tools, find the highest-leverage place to use them, and build the systems or agents that let you get more work done without carrying everything in your own head.