AI · FIELD NOTES ON STRATEGY AND PRODUCT · UPDATED OCT 2026

Most companies use AI. Few can find it in the P&L.

These notes are about the gap: how to redesign work, teams, pricing and quality so AI shows up in the numbers, and the people doing the work get more reach. Think supersuit, not severance package.

44%
of companies scaling AI
37%
see any EBIT gain
6%
high performers who redesign the work

McKinsey, The state of AI in 2026, Aug 2026

NOTES, NEWEST FIRST

NOTE 04P&L

Bolted-on AI doesn't show up in EBIT.

Almost everyone uses AI and few can find it in the numbers. The 6% who can didn't add a chatbot. They redesigned the work. A supersuit, not a severance package: give the people doing the work more reach, then rebuild the process around what they can now do.

37%of companies attribute any EBIT gain to AI; flat year over yearMcKinsey, The state of AI in 2026, Aug 2026

try this · Pick one workflow end to end and redesign it, instead of sprinkling AI across twenty.

NOTE 03ORG DESIGN

The org chart is the next product you'll redesign.

The lines between PM, designer and engineer are blurring. Some product managers now call themselves builders. Fewer people with more reach needs a lot more than a headcount plan: career paths, new rituals, a shared definition of good work.

33%of AI builders expect smaller teams; 45% a different mix of rolesICONIQ State of AI, Jul 2026

try this · Write the new job descriptions before the reorg, not after.

NOTE 02PRICING

Someone is paying for those tokens. Probably you.

AI features cost money every time they run, and most companies are absorbing that quietly. Seat prices hide it until usage grows and margins don't.

13%of software companies fully pass inference costs on to customersNalpeiron, Jun 2026

try this · Instrument cost per feature use now, before finance asks.

NOTE 01QUALITY

Your customers shouldn't be your QA team.

Only one team in five does adversarial testing before release. In any other product we'd call that shipping blind. Evaluation is product work. Put it on the roadmap.

67%of AI teams find problems mainly through user feedbackICONIQ State of AI, Jul 2026

try this · Define what 'good' looks like for your top three AI tasks, then test against it every release.

How I'd run AI in a product org

  1. 01People firstDesign AI around the people doing the work, then the process, then the headcount line.
  2. 02Redesign, don't sprinkleOne workflow rebuilt end to end beats twenty pilots.
  3. 03Measure the moneyCost per use and value per use, from day one.
  4. 04Test before customers doDefine good, then check every release against it.

Putting AI to work in a product org?

I help leadership teams pick the workflows worth redesigning, reshape the roles around them and measure whether it worked.

Email Brandon