The current map
The AI-software-development market in 2026 has two visible poles.
On one end: the big consultancies.Slalom, ICF, Accenture Song, Coherent Solutions. They've added “AI-accelerated engineering” service lines to their pages. Real teams, real expertise, but the operating model is the same as it was in 2018: a discovery phase, a proposal, a partner-led pitch, then a delivery org of 50–500 people who work on your engagement among many. Speed is gated by the headcount math.
On the other end: the AI-pure platforms. Devin, Factory, Blitzy. They sell speed via autonomy: an agent that codes the feature, runs the tests, opens the PR, even merges. Marketing says 5× faster, 80% automated. The demos are real. The architecture choices buried inside the agent's output are also real — and you find out about them six months later when the feature can't be extended because nobody on your team understands the shape of the code.
Both ends are useful. Neither end is what most mid-market buyers actually need.
What's broken about each end
The big-consultancy problemis layering. You buy the partner, you get the senior associate who relays to the senior consultant who manages the consultants who work with the offshore developers. By the time a decision crosses four layers, the sprint is half over. The AI tooling exists in their workflow, but it's wrapped in process. It compresses the typing, not the coordination.
The AI-pure problemis accountability. An autonomous agent doesn't know your business. It doesn't feel the consequence of a bad choice-of-architecture for the next two years of your engineering hires. It optimizes for the issue, not the codebase. When the feature ships and works, that's great. When the feature ships and locks you into a pattern that hurts later, there's no senior engineer to call and ask why.
The shape in the middle
The shape that actually delivers for most mid-market clients is a small senior team — one or two senior engineers, a designer, a PM — paired with the same AI tooling the big firms have and the AI-pure shops claim. The math is:
- Senior engineers carry the architecture. Choices about queues vs crons, auth models, service boundaries, retry strategies — these are made in conversations between the engineer and the client, not in an agent's prompt context. The decisions live for years
- AI tooling carries the typing. Boilerplate, migration scaffolds, test setup, pattern translation, documentation, code search. The pieces of the job that used to take a day shrink to ten minutes. The senior engineer spends those reclaimed hours on the work that actually requires senior judgment
- No middle management. The person you talked to on the fit call is the person typing the code. Decisions move at the speed of one conversation, not four
- One throat to choke.When something ships wrong, you don't troubleshoot an agent. You ask the engineer
The numbers that make it real
A senior engineer working unassisted ships at roughly the pace they did five years ago. Maybe 10% faster from better tooling — autocompletion, linters, etc. Not a step change.
A senior engineer paired with current-gen AI tools (Claude Code, Cursor, agentic test runners) ships 2–3× fasteron the kind of work where AI earns its keep — and the same speed on the work where it doesn't. That averages out to something like a 70% compression on a typical feature.
A team of 8 senior engineers with AI ships approximately what a team of 16 senior engineers shipped two years ago. Which means a small senior team paired with AI replaces a meaningful chunk of a mid-size consulting engagement.
When the middle is wrong
Be honest: the middle isn't always the right call.
If you're a Fortune 500 with a 12-month rollout that touches 200 systems, you want the big consultancy. They're built for the coordination. The slowness is the point.
If you're running a side project and you don't care about the codebase six months from now, the AI-pure platform is the right call. Cheap, fast, throw-away.
If you're a serious mid-market business with a real product that needs to ship in weeks and keep shipping after we're gone — the small senior team plus AI is the shape. That's where ACM lives.
What to ask
When evaluating any of the three shapes, ask the same question and listen for the answer:
“Who is going to type my code, and who is going to defend it when it's wrong?”
The big consultancy says: a consultant, probably offshore, defended by a partner you met in the pitch.
The AI-pure shop says: the agent, defended by your account manager.
The middle says: a senior engineer, defended by themselves. They wrote it, they reviewed it, they understand it, and if it breaks they're the one fixing it.
Pick the answer that lines up with the kind of business you're running.