Golden Data

Issue #14 · July 17, 2026

AI Redundancy Washing Has a Tell.

Paul BrownWrites Golden Data from Ogden, Utah4 min read

Every Friday I look at what moved in the analytics craft. This week what moved was a phrase, and it belongs to a bank’s research desk, not a career coach. Deutsche Bank analysts predicted in January that “AI redundancy washing” would be a defining feature of 2026. Six months in, they were right. More than half of this year’s layoff events name AI as the cause, and Oxford Economics went looking for the productivity gains that would justify the cuts and could not find them at the macro level.

So here is the thing nobody running the layoff tracker wants to say out loud. The tracker is not a labor-market signal. It is an investor-relations document with a body count. If you are a senior analytics IC reading it as a verdict on your craft, you are reading a press release as if it were economics. The cut is real. The stated reason is a narrative for shareholders, and you are not the audience.

What’s actually moving in the market

AI is the named cause of 56% of 2026’s layoff events, and Oxford Economics still can’t find the automation. That is 156,270 workers across roughly 150 companies through H1. Deutsche Bank called it “redundancy washing” back in January; since then Wharton’s Peter Cappelli and Marc Andreessen have said the quiet part, with Andreessen estimating pandemic-era overhiring left large tech firms 25% to 75% overstaffed. One widely-circulated CFO survey this month put planned 2026 cuts citing AI at 502,000, nine times last year’s figure. The cuts are real. The attribution is a costume.

The data analyst role itself has not been automated. SQL and Python have held a top-five spot in required skills for three straight years, even as roughly 92,000 data-adjacent workers took severance over that same stretch. The generalist seat is clearing; the skill is not. Cut and posted-for are happening in the same quarter, frequently at the same company.

Snowflake and Databricks both shipped governance into the semantic layer in June. Snowflake’s Horizon Catalog added Data Governance Skills to Cortex Code — natural-language policy creation and auto-classification. Databricks’ Genie inherits Unity Catalog’s governance posture and reuses its compute. This is not a product note. It is a job-description note: the seat that survives is the one that owns the metric definition and the access policy, not the one that writes the transform.

Senior Analytics Engineer comp sits at a $183K average, a $153K–$222K interquartile range, and $263K at the 90th percentile (Glassdoor, July 2026). The number at the top of that range attaches to end-to-end ownership — semantic layer, lineage, governance — not to the volume of dashboards shipped. The band is wide because the market is repricing what the title means in real time.

What I’d do this week

Stop reading the layoff tracker. Read the postings from the companies on it. The tracker tells you a company cut roles and blamed AI. Its careers page tells you what it is paying for right now. When those two disagree — and they almost always do — the postings are the truer signal, because a job req is a budget commitment and a layoff press release is a story told to investors.

  • The user moment: The next time you catch yourself scrolling a layoffs tracker at the end of a long day. Close it. Open three careers pages instead.

  • The shape: A one-page divergence note.

  • The time budget: 30 minutes.

  • The artifact: For three companies that announced AI-cited cuts this year, the exact phrases in their current Senior/Staff AE postings — “Unity Catalog,” “Cortex Analyst,” “semantic view,” “governance,” “evaluation.” The list of what they cut, next to the list of what they are hiring for.

  • What success looks like: A month of this and you stop reading headlines as verdicts on your worth. You read them as what they are, and you know precisely which competencies the same companies are still writing checks for.

Put one governed metric into your semantic layer this week. The postings name a specific seat: the person who owns the metric definition, its lineage, and its access policy — the seat an agent cannot occupy without a human’s judgment behind it. You do not earn that seat by reading about it. You earn it one metric at a time.

  • The user moment: Wednesday, in the block you would otherwise spend rebuilding a dashboard someone changes their mind about by Friday.

  • The shape: One committed semantic-layer definition with a governance note.

  • The time budget: 90 minutes.

  • The artifact: A single contested metric — “active user,” “net revenue,” something two teams argue about — defined in your semantic layer (a dbt semantic model, a Cortex Analyst semantic view, a Unity Catalog metric), with an owner, a lineage note, and one access rule. Committed, not messaged.

  • What success looks like: Four governed metrics by month’s end, and a real answer to the question the market is now asking in interviews — “what do you own that an AI agent can’t own without you.” Most senior ICs answer that with a job title. You would answer it with a commit history.

Sources

Crafting runs weekly. The job is to be useful, not to sell. If a line in here is close to a move you're working on, the email reaches a person.

Hiring rather than hunting? Signal is the column for your side of the desk.

Email meRead more Crafting

← All Crafting issues · Drafted with Claude · Edited by Paul Brown