Amazon has published a new guideline on the role of the software developer in the age of AI, and one line in it is doing more work than its authors may have planned: if your work does not benefit from current AI tooling, “that is a finding, not a failing.”
The document, written by Terk Terkowitz — whom The Register describes as a VP and Distinguished Engineer at the top of Amazon’s individual-contributor technical track — was circulated internally in September and published publicly on Amazon’s site on 27 October. The Register’s Corey Quinn picked it up on 31 October, reading it as an attempt to steady a workforce whose mood “feels mighty bleak.”
🔍 The Bottom Line
The memo’s single most career-relevant instruction: an engineer who documents where AI does not help their work is demonstrating the judgement Amazon says it wants. Whether that holds under real promotion decisions is the open question The Register raises — and it is the question every NZ knowledge worker should be asking of their own employer’s AI mandates.
What the Memo Actually Says
The guideline asks engineers to bring evidence rather than vibes to the AI-transition. Its concrete instructions, from Amazon’s published guideline:
- If AI tooling does not benefit your work, document why, share it with peers, and take it to your manager — “that is a finding, not a failing.” An engineer who can explain where AI does not yet help “is demonstrating the judgement we are asking for.”
- Input metrics like lines of code or token counts are “deeply flawed” and could drive negative outcomes if used as targets — “tokenmaxxing is at odds with frugality.”
- Teams and individuals “not adopting AI or increasing output metrics are falling behind.”
- Use AI adoption metrics “with care and judgement,” paying attention to the 10th percentile of adoption (teams not embracing AI, and why) and the 90th percentile (whether heavy AI use actually raises delivery).
- Leaders should have used the tools on production work and be able to name a task where they helped and a task where they did not.
The Contradiction The Register Spots
The Register’s critique is precise: the memo concedes input metrics are deeply flawed, then says to use them anyway, then warns that non-adopters are falling behind — three positions The Register calls unaligned.
Quinn’s test for whether any AI mandate is well-considered: what happens to an engineer who documents that AI made a task worse. Taken seriously, the mandate has substance. Marginalised or managed out, it’s cover — a manager can cite “a finding, not a failing” for a strong engineer who skips the tool, and “falling behind” at PIP time, with both readings congruent with the written policy.
The Register also reports that Amazon announced around 250 further job cuts the same week. It cannot say whether the memo and the cuts are related — “I can’t tell you that these things are related,” Quinn writes — but notes that every engineer reading the memo assumes they are.
Tokenmaxxing Shows Why the Fear Is Rational
The nervousness has a track record. As CNBC reported on 27 October, Amazon earlier this year shut down an employee-built leaderboard that ranked engineers by AI token usage, after workers gamed it — spinning up junk tasks to climb the ranks. The practice earned the name “tokenmaxxing,” and the new guideline itself calls it “at odds with frugality.” The measuring-costs problem has since spread past engineering — Atlassian added AI wallets and tokenmaxxing guardrails to corporate spending as AI budgets balloon. Meta ran a similar token leaderboard across its 85,000 employees, per The Register.
The pattern generalises into career advice grounded in the data: when a workplace measures tool activity rather than outcomes, workers rationally optimise for the metric, and the measurement becomes the job. A worker’s safest position is documented outcomes — where AI demonstrably helped, where it demonstrably did not, and what the delivery numbers show. It also cuts both ways: firms that cut staff for AI are quietly rehiring when the tools underperform — the “AI boomerang” reported by AP, with 32 percent of AI-motivated cuts later rehired per Robert Half.
The NZ Angle: Exposure Without a Playbook
New Zealand workers are inside this same shift, and the country has an unusually specific map of where. A Reserve Bank analytical note published in February used ten different large language models to score AI and robotics exposure across New Zealand occupations, finding AI exposure “most strongly associated with professional, managerial, and administrative occupations” and with middle to upper-middle income brackets between $50,000 and $150,000.
The same note found about 30 percent of NZ workers face high combined AI-and-robotics exposure, while fewer than 5 percent face low exposure to both. Few NZ occupations are positioned to avoid engaging with at least one of these technologies — which makes the question of how employers treat AI scepticism a broad one, not a Silicon Valley curiosity.
Site analysis of NZ job ads has tracked AI-skills demand running well ahead of broader listings — SEEK NZ data reported July AI-skills postings doubling year-on-year, and SEEK’s September report recorded a 42 percent annual rise even as total ads fell. The premium on demonstrable AI skills is also well documented: PwC’s AI Jobs Barometer analysis of labour market data across 15 economies finds workers with AI skills command a 56 percent wage premium over comparable roles — evidence that visible, verified AI capability is what the market currently pays for. Documented judgement about where AI doesn’t work is the same skill from the other side. It is also exactly what the RBNZ’s graduate-ladder analysis warns is disappearing when entry-level rungs collapse — judgement has to be learned somewhere.
The Career Question Worth Asking
The memo contains genuinely reasonable guidance — The Register calls it “a good document” — including its instruction to check whether the heaviest AI-using teams actually deliver more. The open issue is the gap between policy language and management practice, as Quinn’s promotion-question underscores.
For now, the practical read for NZ workers: if an employer’s AI mandate comes with metrics, ask which percentile you are being measured in, keep a private record of where the tools helped and where they did not, and note everything in writing. Amazon’s own document says documented non-use is a finding. If a workplace punishes exactly that, workers have learned something the policy would never admit.
FAQ
What does Amazon’s “a finding, not a failing” mean? It is a line from Amazon’s engineering guideline saying that if your work genuinely doesn’t benefit from current AI tools, documenting that conclusion is treated as useful engineering judgement rather than resistance. The Register asks whether promotion and performance decisions will actually reward it.
What is tokenmaxxing? Burning AI tokens to inflate usage metrics — spinning up junk tasks to climb an internal leaderboard. Amazon shut its token leaderboard after the gaming became obvious, per CNBC, and its new guideline calls tokenmaxxing “at odds with frugality.”
Does this apply to New Zealand workplaces? The memo is Amazon-specific, but the measurement question is general. RBNZ’s exposure analysis found most NZ occupations face meaningful AI or robotics exposure, so how employers measure AI use — and how they treat documented non-use — matters well beyond Amazon.
Is Amazon laying off engineers right now? The Register reports around 250 further job cuts announced in the same week as the memo’s coverage, and says it cannot establish a connection between the two. Amazon has also cut larger numbers earlier in 2026.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
