Sergey Tikhonchuk

AI + OKR = Strategic AI

[Sergey Tikhonchuk @stikhonchuk]

Her own CEO asked managers to picture "autonomous AI agents already part of the team." The X-shaped person is who manages that team. Tools are the easy half; judgment and process view are the product.

[Sergey Tikhonchuk @stikhonchuk]

You can't hold that view from inside one deep specialty. A T-specialist owns one agent in their domain and is blind to the rest. The X-shape is the shape of the layer that manages a bench of them.

[Sergey Tikhonchuk @stikhonchuk]

From the build side, that X-shaped person has a job. When an agent does the narrow execution, what's left for the human is judgment on the cases it gets wrong — and the view across the whole process.

[Sergey Tikhonchuk @stikhonchuk]

Shopify's COO says AI is retiring the "T-shaped" employee — broad skills, one deep specialty — for "X-shaped" people: several spikes of competence, "a seven out of 10 designer if you want to be."

[Sergey Tikhonchuk @stikhonchuk]

Then ask it something the handbook doesn't contain. It won't invent one. The ceiling on this kind of assistant was never the model — it's what your company has actually written down.

[Sergey Tikhonchuk @stikhonchuk]

We put a demo online rather than describing it. Fictional company, real handbook, assistant in the portal and the same one in Telegram. Ask it something that isn't on the checklist.

[Sergey Tikhonchuk @stikhonchuk]

An answer that varies by who you asked isn't a policy at all.

[Sergey Tikhonchuk @stikhonchuk]

Every new hire asks roughly the same fifteen questions. The cost isn't the minutes — it's that three people answer, each recalling the rule instead of reading it, and the new hire gets three versions of one policy.

[Sergey Tikhonchuk @stikhonchuk]

Better unit than payroll: an agent closing 80% of a process leaves the human touching 1 case in 5. Not 0.8 of a person removed — the same attention over 5× the volume.

[Sergey Tikhonchuk @stikhonchuk]

So: announce AI → cut staff → survivors must adopt a tool they just watched remove colleagues → the gain you paid for depends on that adoption. The cut suppresses the payoff.

[Sergey Tikhonchuk @stikhonchuk]

Employee sentiment toward AI strongly tracks firm productivity. The biggest driver of negative sentiment, by far: fear about job security.

[Sergey Tikhonchuk @stikhonchuk]

Ma et al. (Univ. of Pittsburgh) matched millions of Glassdoor reviews against hundreds of AI-investment and layoff announcements, 5 years of US public companies.

[Sergey Tikhonchuk @stikhonchuk]

Nearly 3 in 10 CEOs say cutting headcount is the main reason they're investing in AI. New research suggests they're the ones least likely to see a return. 🧵

[Sergey Tikhonchuk @stikhonchuk]

Full piece 👇
https://t.co/I72p66po8N

[Sergey Tikhonchuk @stikhonchuk]

The filter most teams skip when choosing that process:

Pick the one whose data the next project can reuse. Same customers, same records.

Sequence by adjacency, not by which win is easiest — or you get three working tools that share nothing.

[Sergey Tikhonchuk @stikhonchuk]

The practical consequence is smaller than either implies.

No enterprise data warehouse. A data foundation: enough clean, governed data about one painful, well-defined process.

A mid-sized company can finish that.

[Sergey Tikhonchuk @stikhonchuk]

Which is where both standard recommendations fail.

"Finish the transformation first" treats the discipline as a wall to clear before starting.
"Our AI ingests messy data" claims it's handled for you.

It was never a wall. It's the part that stays.

[Sergey Tikhonchuk @stikhonchuk]

The discipline does not move at all.

Point a capable model at an ungoverned mess and you don't get insight. You get fluent, authoritative answers built on contradictions the model cannot see — expensive precisely because they read as correct.

[Sergey Tikhonchuk @stikhonchuk]

The labor genuinely drops.

Deduplication. Reconciliation that consumed analyst-weeks. The matching that spots "J. Smith Ltd" and "John Smith Limited" as one customer across two systems. Monthly reports that can now run continuously.

[Sergey Tikhonchuk @stikhonchuk]

AI moves the labor of data work. It does not move the discipline.

Most of the disagreement about cleaning your data before adopting AI dissolves once you separate those two.