Will AI replace our developers?

It will replace the part of their day spent on boilerplate, and it already has. It will not replace the part spent understanding a client's process well enough to know which part of it to automate. Agencies pivoting successfully move engineers toward that second part and let the models do more of the first.

Do we need to rebrand?

Not at first. A rename signals a change of focus, and it is worth doing once the focus has actually changed. Several agencies on this site kept their name and added a diagnostic to the front of every engagement; a few renamed after the old work had stopped. Rename once the pipeline agrees with the new name.

How do we price the first AI engagements?

Lead with a fixed-price discovery, a week or two, with a written roadmap and costed options as the deliverable. Price it low enough to be an easy yes. Then price the builds against the outcome the roadmap promises. Your old hourly rate is a floor for sanity checks, and a poor way to sell work whose value is measured in hours saved.

How to pivot a dev agency to AI

A development agency pivots to AI by changing what it sells first. The old proposal opens with a build and a rate. The new one opens with a paid diagnosis that ends in a ranked list of what to automate and what each fix is worth. The engineers stay. The pitch changes. The old repositories become an asset.

What you already have

A dev agency that has shipped for a decade holds three things a new AI agency would kill for. Process knowledge: you have seen inside a hundred operations and you know which handoffs leak hours. Delivery discipline: staging environments, version control, testing, the boring competence that separates a production system from a demo. And clients who already trust you with their systems and will listen when you tell them what to fix next.

What you probably lack is the habit of charging for the first of those. Most dev shops give the diagnosis away in a sales call and bill for the build. The pivot is largely the act of reversing that.

What changes

The front of the engagement. Sell discovery as a product with a price, a duration and a deliverable. A week is enough for a mid-market firm: process maps, a costed roadmap, a prototype of the first candidate. The client leaves owning something whatever they decide next, which is why it sells.

What you build. Less of it, and more of it on orchestration tools. Much of what used to need a custom application is now an n8n workflow with a model in the middle. Reserve custom code for the cases where a workflow tool will not carry the weight, and say so in the proposal, because clients now ask.

How you charge. Discovery at a fixed price. Builds against the outcome, with your old hourly rate as the floor for a sanity check. A retainer for everything after go-live, described in plain words. The retainer is where a dev shop’s margin has always been thinnest and where an automation practice’s is thickest.

What to stop

Stop taking general development work, or at least stop advertising it, because a services page that lists AI beside mobile apps and WordPress tells the buyer you are a generalist and prices you as one. Stop bidding on specifications. The client who has already written the spec has already decided what to build, and the value you add now is upstream of that decision.

This is the hard part of the pivot, because the old work still pays. Most agencies that manage it run both for a year and let the new work crowd out the old, rather than cutting revenue on a principle.

The old clients

They are the first pipeline. Every one of them has a process you already understand and an operation you already have access to. A short note offering a paid diagnosis of the process around the system you built converts well, because you are the only vendor who can make that offer with any credibility.

The old repositories

A dev shop with ten years behind it owns more code than it thinks. Internal tools. R&D that never shipped. Products that were shelved. Client projects where the contract, read carefully, left the IP with you. Most of it sits on a drive earning nothing.

There is now a buyer. AI labs license private codebases, source and git history both, to train and evaluate their coding models. Licenses are non-exclusive, so a repository can be licensed more than once and you keep full ownership and use. The published rates, checked on 15 September 2026, are indicative at $100 per million tokens of source and $40 per million tokens of history, and the history is where most of the tokens are. Only code you own qualifies, and client-owned work is excluded outright, which makes sorting what you own the first job.

The mistakes we see most

Rebranding before the work has changed, so the site promises what the pipeline cannot deliver. Keeping the hourly model and wondering why automation projects feel small. Hiring an “AI person” instead of teaching the engineers who already know the clients. And building the first agent for a client who needed a workflow, because agents are more interesting.

Once the first three engagements have numbers attached, list the agency. Buyers on this site are looking for exactly the combination a pivoted dev shop has: delivery discipline and a named client list.