Standout Skills Self-Review
The candidates every leader agreed on owned the consequence. Rate yourself on evidence, not intent.
Built by Sunil Mathew, co-authored with Claude
- What to bring
- One project you worked on in the last 6–12 months; its goal and what actually happened; links to the work you’d point to (design docs, reviews, evals, incident notes, decisions); and the name of one person who worked with you and could confirm your answers.
- What to do
- Log your evidence from that project, rate yourself on 16 statements in four groups, mark which ratings you can back with evidence and which someone else would confirm, then pick one 90-day action and one longer-term action.
- What you leave with
- A score for each of the four groups, a profile (Standout pattern, Building or Close this gap first), a list of the ratings you claimed but can’t yet evidence, and a two-action plan with a review date.
Excel workbook · also opens in Google Sheets and LibreOffice · free
From The Living Craft, practical learning and consulting with Sunil Mathew. Explore the programmes
Every leader in the room agreed. The question is what they saw.
Sunil has taken part in performance and promotion reviews for senior engineers, managers and architects at Google, Amazon and Walmart: intense, deliberate evaluations. Some candidates stood out, and every leader in the room agreed. They owned the consequence, and showed technical depth and sound judgement that gave them a disproportionate impact on, and influence over, the outcome of the project or product. As agents write more of the code, Andrew Ng’s AI Engineering Skills Map describes engineering work as ‘shifting toward deciding what should be in the spec.’ This checklist puts the two together.
Three traits, and the result they produced.
| Group | What it means | What it looks like in the evidence |
|---|---|---|
| A · Owns the consequence | Cares about the outcome, not just the delivery | Knows whether the work changed what it was meant to change; leads the fix when it breaks after launch |
| B · Technical depth | Understands the system well enough to see what others and agents miss | Can explain failure modes; catches a design trade-off an agent got wrong |
| C · Sound judgement | Makes the right call under uncertainty, and can explain it | Decides what goes in the spec; knows when to ship fast and when to slow down |
| D · Influence on the outcome | The traits above change decisions beyond their own work | Other teams’ decisions change because of what they wrote or showed; leaders ask for their view early |
How it links to Andrew Ng’s skill map
The AI Engineering Skills Map names four skill areas that, it says, all developers will need:
- Building and deploying AI applications
- Software engineering fundamentals
- Using coding agents
- Shaping the build
Two statements link to evaluation and error analysis, which the letter describes as a core skill within building and deploying AI applications.
The link from each statement to a skill area is Sunil’s interpretation, not Ng’s.
Not affiliated with or endorsed by Andrew Ng or DeepLearning.AI.
The three questions
| # | Question | Tab |
|---|---|---|
| 1 | What’s my evidence from one recent project? | 1 · Evidence Log |
| 2 | Which statements can I rate “Consistently”, with evidence? | 2 · Self-Review |
| 3 | Which skill do I build next, and how? | 3 · Plan |
One question each, and the mistake it catches.
What’s my evidence from one recent project?
The mistake it catches Rating yourself on intent instead of evidence. You write down what you did and what it changed before you rate anything, so each rating has something to point to.
Which statements can I rate “Consistently”, with evidence?
The mistake it catches Claiming “Consistently” for something nobody could confirm. A “Consistently” with no evidence counts as “Sometimes”, and the sheet counts the ones nobody has confirmed yet.
Which skill do I build next, and how?
The mistake it catches Trying to improve everything at once instead of the one gap that holds the rest back. The plan starts from your lowest group and its weakest statement.
Rate each one Not yet, Sometimes or Consistently. Then say what shows it.
A “Consistently” with no evidence counts as “Sometimes”. This is a self-review, not a promotion predictor. Use it to choose what to build next.
A · Owns the consequence
- A1
For my last project, I can say what outcome it was meant to change and whether it did, with a number or an observation.
- A2
When something I shipped went wrong after launch, I led the fix and the follow-up, not just my part.
- A3
Before building, I write down what would prove the approach wrong, and I check it afterwards.
- A4
I measure the output of the agents or AI features I own after release, using evals or error analysis, not only before release.
B · Technical depth
- B1
I can explain the failure modes of the system I own: what breaks under load, on retries or with bad data, and how it recovers.
- B2
I can review agent-generated code and spot a design trade-off the agent got wrong.
- B3
I give coding agents precise context, and I know when to stop and intervene.
- B4
I have built and deployed at least one AI feature end to end, including how it’s evaluated.
C · Sound judgement
- C1
I decide what goes in the spec, including acceptance criteria and what’s out of scope, before anyone writes code or prompts an agent.
- C2
I know when to ship a quick MVP and when to slow down, and I can explain the call.
- C3
I ground technical decisions in the customer’s goal and the business context, not only in technical elegance.
- C4
I change my mind when the evidence contradicts me, and I record why.
D · Influence on the outcome
- D1
Another team’s decision changed because of something I wrote or showed.
- D2
I’ve raised the bar for others: reviews, standards or tools that other people now use.
- D3
I’ve helped a less experienced engineer build judgement, not just get answers, through reviews, evals or error analysis.
- D4
Leaders ask for my view before a consequential decision, not after it.
A senior engineer, one recent project.
Illustrative example: a senior engineer reviewing one recent project. Not a real person, company or real figures. It is filled in on the workbook’s last tab.
16 statements
One “Consistently” (B3) had no evidence, so it counted as “Sometimes”. Total 16 of 32 counted, 17 claimed.
2 of 8
D · Influence on the outcome, so the profile was “Close this gap first”.
D1
Another team’s decision changed because of something I wrote or showed.
Every leader in the room agreed on the ones who owned the consequence. Start there.
Practise this judgement on a real agentic system. Explore Architecting Agentic AI Systems: 30 live hours with Sunil Mathew; confirm complete setup and pre-session preparation before joining.
Practise this judgement on a real agentic system. Explore Architecting Agentic AI Systems: 30 live hours with Sunil Mathew; confirm complete setup and pre-session preparation before joining. Request a fit call at https://learning.thelivingcraft.ai/programmes/agentic-systems/architecting-agentic-ai-systems/#discuss-architecting-agentic-ai-systems
Published 20 October 2026 · Content CC BY 4.0 · © The Living Craft