The way most leaders evaluate a development partner — team size, hourly rate, velocity metrics — was built for a world that no longer exists. AI has broken the link between developer hours and delivered value, and it broke it unpredictably. If you're applying 2018 criteria to a 2026 vendor, you're going to misallocate budget and be disappointed by the results.

The shift is already visible. Teams are shipping the same or more with fewer people. One CTO cut his team from 12 engineers to 8 and shipped 30% more features the next year. Google restructured 30-person teams into groups of 3–6 and improved throughput. Headcount stopped being a proxy for capability.

Five things to actually evaluate

1. Specific AI integration. Not "AI-augmented delivery." Real orchestration patterns, review gates, and a straight answer about how senior engineers redeploy the time AI gives back.
2. Stated refusals. Strong partners have boundaries — they'll decline unclear ownership, velocity-without-judgment contracts, and "AI-only" work. What a vendor won't do tells you more than what they will.
3. Senior coverage, honestly. Who really has 10+ years of operational experience? And what happens to the engagement if that person leaves?
4. Foundation discipline. CI/CD, testing, code review, deployment safety. This is the difference between AI being a 5x multiplier and AI being the reason production goes down at 2am.
5. Evidence of judgment. Ask about mistakes and post-mortems. A clean, unbroken success narrative is a red flag, not a green one.

Red flags

Vague "AI-augmented" claims with no architecture behind them. Headcount presented as capability. Fixed hourly rates against undefined scope. Feature-velocity promises with no willingness to say no. An inability to describe their own failure modes.

The pricing tell

Hourly billing prices effort in a world where value has become outcome-dependent. A senior engineer with a well-instrumented AI workflow can produce in a day what a junior produced in a week three years ago — so what exactly is the hour measuring?

A retainer model tends to align better: it preserves continuity of context, points incentives at outcomes instead of billable hours, and lets you have strategic conversations without a meter running. It's not universal — it's a poor fit for true one-offs, body-shop staffing, or engagements where "success" was never defined. But when the work is ongoing and the outcome matters, it's the honest structure.

The real bottleneck here isn't vendor capability. It's the buyer's mental model. Update the questions you ask, and you'll be shocked how quickly the right partners separate from the wrong ones.