Fundamentals first
The engineer reads, navigates and debugs the codebase without AI. If they cannot reason about the system, they cannot judge when an agent is wrong.
Engineering changed. Our hiring standard did too.
The best engineers no longer win by writing the most code. They turn vague tickets into shipped changes they can vouch for, using agents to compress the typing while keeping ownership of every decision.
See our hiring standardThe candidate journey
Six stages. One consistent standard.
Each stage answers a different question. Together they show whether someone has the craft, communication and judgement to perform in a modern engineering team.
- 01
Application review
We look for evidence of shipped work, depth in the relevant stack and a clear match with the role — not keyword density.
- 02
Introductory conversation
We assess communication, working preferences and whether the candidate can reason beyond prepared answers.
- 03
Technical fundamentals
The engineer works in a real repository without an agent. We observe how they read a system, prioritise and debug.
- 04
AI-native assessment
In the same environment, AI tools are switched on. We assess how the engineer directs, challenges and verifies their agents.
- 05
Leadership and references
For senior and client-facing roles, we validate professional judgement, consistency and the ability to work well with others.
- 06
Evidence-based decision
The complete set of observations is considered together. No single impression is allowed to outweigh the evidence.
The technical standard
Same engineer. Same repository. Two different tests.
Judgement at speed
AI tools are introduced in the same environment. We watch how work is planned, delegated, challenged and verified — not just the output at the end.
“An engineer can only ship what they can verify. Judgement is the limit now, not output.”
The AI-native scorecard
Six dimensions we can observe.
“AI-native” is not a label on a CV. It is a set of working behaviours that become visible when an engineer has to deliver inside a real system.
AI output verification
Catches hallucinated APIs, subtly wrong logic or security gaps in generated code before it ships, rather than trusting it because it compiles.
Tool orchestration
Knows when to hand work to an agent and when to hand-code, and can run both in parallel without losing control of the work.
Spec-driven development
Breaks a vague ticket into pieces small enough for the tool to execute well, rather than throwing the whole ambiguous ask at it in one go.
Prompt and context engineering
Sets up the prompt and task context so the tool produces usable output in one or two passes, instead of repeating the same vague ask.
AI-accelerated debugging
Uses the agent for root-cause analysis, log triage and stack traces without skipping reproduction or understanding the failure.
Context-switching hygiene
Knows what each agent is doing, returns to the right one and does not let notifications drive the switching.
What good looks like
We stopped judging the output.
We started judging how it was produced. Writing code has become cheap; understanding how a system fits together hasn't — if the fundamentals aren't there, the candidate has nothing to check the agent against.
- Reads what the agent produces and pushes back on it
- Sets up worktrees and subagents so several tasks run at once
- Plans before building — each PR small enough to read in one sitting
- Gets usable output in two or three tries, with scope and context set up front
- Reproduces the bug and understands the cause before fixing it
- Keeps track of four or five parallel streams without losing one
Miyagami whitepaper
The Next 10X Engineer
Why the best engineers now write less code, how to tell them apart from everyone else, and what AI-native engineering changes about hiring and team design.
Download the whitepaper