The two engineering profiles we now hire for.
Raw output is not the bar anymore. We index on creative builders with product sense, and deep systems experts for the parts where subtly wrong is still wrong. Here's how we screen for both inside enterprise teams.
When we help an enterprise team reshape itself around AI-native engineering, hiring is the lever that compounds the longest. The profile that won the last decade — the senior IC who can type fast, hold a large codebase in their head, and ship clean PRs — is still valuable, but it is no longer the scarcest profile.
Two new profiles have taken the top of the funnel. We index on both.
Profile 01 — Creative builders with product sense
The engineer who can spot the right thing to build and prototype it fast. Taste is scarce. Typing is not. This profile thrives in the new economics: they convert an ambiguous problem statement into a working artifact in a day, then iterate against real feedback.
What we screen for:
- A portfolio of shipped weekend projects, not just employer logos
- Strong opinions about UX and product surface — they can explain why a flow is wrong
- Comfort working from a vague prompt instead of a JIRA ticket
- Evidence they cut scope ruthlessly to ship
- Fluency with agent tooling as a native medium, not a curiosity
The failure mode of this profile is shipping fast but shallow. We pair them with the second profile to keep the bar honest.
Profile 02 — Deep systems experts for the hard parts
The places where trust-but-verify matters most: distributed systems, data integrity, auth, payments, cryptography, kernel code, latency-critical paths, anything where subtly wrong is still wrong. Agents are competent here but unreliable. A human who can read the diff and catch the off-by-one in the consensus algorithm is non-negotiable.
What we screen for:
- Real production scars — paged at 3am, root-caused, shipped a fix
- Comfort reading code in languages they didn't write
- Ability to articulate failure modes before they happen
- A strong default toward "let me verify that claim" rather than "the model said so"
- Patience to do the boring work of writing the spec and the proof
What we index on less
We don't care how many lines you can write per hour. We care what you choose to build and how you know it's right.
Raw output volume has collapsed as a signal. Two candidates with identical line counts can have wildly different impact in 2026. The interesting signal is judgment — what did they choose not to build, what did they catch in review, what did they refuse to ship without verification.
How the two profiles compose inside a pod
Our current default for an embedded engagement is a pod of three: one Profile 01, one Profile 02, and one senior FDE to set the bar and own the handover. The Profile 01 engineer runs the agent fleet and ships the surface area. The Profile 02 engineer guards the trust boundaries and reviews anything that touches money, identity, or customer data. The FDE owns the architecture and the relationship.
That shape consistently out-ships pods built on the old "five senior generalists" model — and it costs less.
The hire-for-future test
If you are staffing a team in 2026, the question we ask hiring managers is simple: does this candidate make the team faster at choosing, or faster at typing? If it's the latter, you are hiring for a constraint that no longer exists. Hire for the new one.