The question usually arrives politely disguised — “so, how are you thinking about AI?” — but what buyers actually mean is sharper: am I about to sign a contract for a team that the tooling makes obsolete? It is a fair question, and it deserves a straight answer rather than a defensive one. Something real is changing about the economics of software teams. It is just not the thing the headline suggests.
What AI genuinely compresses
Agentic tooling is genuinely good at work that is pattern-dense and judgement-light: database migrations from a schema diff, scaffolding for a service that mirrors the last one, API clients generated from a spec, test fixtures, mechanical refactors across a hundred files, the fourth CRUD endpoint that looks like the first three. We run this daily inside a trunk-based pipeline and published an honest inventory of the split in what changed in our review gates. The short version: the mechanical middle of a change collapses. That middle was a real share of every team’s billable hours — and any vendor whose value proposition was typing speed now has a problem.
What it does not compress
Judgement. Which trade-off the domain can live with. Which feature not to build. Whether this dependency will still be maintained in three years. Where the security posture needs to be paranoid rather than conventional. Architecture is made of these calls, and they live in context the tooling does not have — conversations with your product people, scars from the last incident, knowledge of where the roadmap is going.
And accountability. When something breaks at 2am, the question is who understood this change, and the answer has to be a person. In our pipeline an engineer signs every merge, and the signature means something specific: I can explain every line of this diff. The same bar for machine-written code as for ours. None of this got cheaper when the tooling arrived. It got scarcer, relative to everything around it.
The economics shift — but not the way the headline says
The naive reading is: fewer hours per feature, therefore fewer engineers, therefore outsourcing shrinks. The actual mechanics run differently. Fewer hours per feature does not mean fewer seniors — it means the same seniors ship more, because the hours AI removes are precisely the ones seniors resented spending. The winning configuration is senior engineers, AI-assisted, with review protected as the scarcest step in the pipeline.
The configuration it undercuts is the opposite one: large, cheap, manual. A team whose margin depended on billing many junior hours against mechanical work is selling exactly the thing that no longer needs many humans. Cheap-plus-manual does not compete with senior-plus-assisted. It competes with a subscription.
What buyers should demand from any vendor now
Two questions, and you need good answers to both:
- Show me your AI workflow — where does the tooling generate, how is work specified for it, and what kinds of change go through it? A vendor with no answer is billing you human hours for machine work.
- Show me your review gates — what stands between generated code and main? Static analysis, eval suites where behaviour is AI-facing, and a named engineer’s sign-off on every merge. A vendor with tooling but no gates is shipping unreviewed output into your product at speed.
Either answer alone is a red flag. Together they tell you whether you are buying a delivery system or a demo. They belong on the same checklist as the questions in our guide to evaluating offshore development quality — and they are quickly becoming the most important items on it.
Who AI actually replaces
So: will AI replace outsourced developers? It is already replacing a specific kind — the bottom-dollar body shop whose entire offer was labour arbitrage on mechanical work. The model does the mechanical work now. What it cannot replace is a senior team that owns architecture, signs its merges, and uses the tooling as leverage rather than as the product. The rate-card comparison that used to decide these deals was always misleading; now it is measuring the part of the work that is evaporating. The question worth asking a vendor is no longer “how much per hour” but “what do your seniors do with the hours the machine gives back”.
This split — AI on the mechanical, seniors on the decisions, gates on everything — is how our Applied AI & Automation practice works, on client systems and on our own pipeline alike. If you are weighing what AI means for a team you are about to hire, bring the question to a technical call, not a sales call.