Can you use AI tools instead of a consultant?

Yes, for a real and growing set of jobs. The tools available to a curious person with a free afternoon are substantially better than what a firm would have charged for a few years ago, and pretending otherwise would be silly.

So here is an honest map of where the do-it-yourself route wins, and where it stops.

Where doing it yourself clearly wins

Anything one person does alone. Drafting, summarizing, rewriting, cleaning up a list, extracting structure from messy text, first-pass research. If the work begins and ends with one human at one screen, a subscription and an hour of practice beats any engagement.

Learning what the technology can do. You cannot outsource intuition. A month of somebody senior actually using these tools daily will do more for your judgment than any presentation, and it makes you a much harder client to mislead.

Small internal glue. Modern automation platforms and a patient person can connect two systems and move information between them. Not elegant, occasionally fragile, and genuinely sufficient for a lot of jobs.

Finding out whether an idea is worth anything. Do the process by hand with a chat tool for two weeks before commissioning anything. If it does not help when you are doing it manually, automating it will not rescue it. This is the cheapest test in the entire field and almost nobody runs it.

If your situation is on that list, spend the money on a subscription and someone's time instead.

Where it reliably stalls

Four walls, and they arrive in a predictable order.

Your data does not agree with itself. The moment a job needs information from more than one system, somebody has to decide which record is right when the customer's name is spelled two ways and the identifiers do not match. This is where most self-built projects quietly die, and it is not a tooling problem, so no better tool fixes it.

Nothing is reliable enough to leave alone. A tool that works when you are watching is a demo. Something touching customers needs to handle the case nobody anticipated, know when to stop, escalate to a person, and fail safely. That gap between working and dependable is where the real engineering is, and it is much larger than it looks from the demo.

It has to run when nobody is thinking about it. Self-built automations depend on the person who built them. They break when a system updates, and the person who understood it is on vacation or has left. Something running unattended needs monitoring and an owner.

You cannot see the whole picture from inside. The genuine limit. You are in the operation, and the mental model you have of it stopped matching reality at some point. Ranking your own opportunities is the thing that is hardest to do for yourself, which is a different service from building.

The middle path most companies should take

Do the cheap experiments yourself, first, for a couple of months. Get a feel for what these things are good at. Kill three ideas cheaply.

Then, if you still cannot rank what to do next, buy a diagnosis rather than a build. That is a smaller purchase than a project and it is the part you cannot do for yourself.

Then build only what survives that, and pay somebody for the piece that has to be dependable rather than the piece that has to be clever.

The self-serve trap worth naming

There is a specific failure that costs more than doing nothing: the subscription that becomes shelfware.

Somebody buys seats for the team with real enthusiasm, there is a training session, and by week three usage has collapsed to two people. Now you have a recurring cost, a reasonable belief that this does not work for you, and no diagnosis of why. That dead pilot then becomes an argument against trying again for a year.

The pattern is almost always that the tool was bought before anybody identified the job. Job first, tool second, in that order, every time.

If the answer turns out to be that your people should own this permanently rather than a supplier, the training comparison is the next page.

Our honest position

We sell consulting and we build agents, and we would still tell a company with the profile at the top of this page to go and try things themselves for two months. It costs them very little, it makes them better at judging vendors including us, and if they come back the conversation is far more productive because they have opinions grounded in use.

Where we think outside help earns its place is the ranking problem and the dependability problem: knowing what to do first, and building something that runs unattended and behaves correctly at three in the morning. That is what the engagement covers.

If you want to test your own readiness before talking to anybody, the question at the top of this page is the one that settles it.