Oabo Insights

On Q2 2026 earnings calls, most of the S&P 500 talked about AI and almost none could put a number on what it returned. That gap is not modesty — it is a missing record.

Enterprises swung from encouraging maximum AI use to clamping spend in a quarter. Both moves were made without an outcome record — and a meter cannot supply one.

A capability improvement can change task design, review rates, cost shape and the business case. The old estimate is not automatically still true.

Frontier or open-weight, the choice shapes cost, control and recovery — not just token price.

The business case for open-weight models is often operational choice, deployment control and switching flexibility—not a promised lower cost.

Code generation can create engineering capacity. Product value appears later—after review, release, adoption and an observed business consequence.

Capacity gained in automotive engineering matters when it changes a program gate, prototype spend, validation load, warranty exposure or another observed business event.

Long-cycle scientific outcomes are important. The nearer proof can be an earlier, better-supported decision not to release more committed spend.

Clicks and attributed revenue can be useful signals. They are not the same thing as recognised margin after discounts, returns, fulfilment and cannibalisation.

Documentation time is a capacity signal. It becomes an enterprise outcome only when the operating system actually changes around it.

For asset managers, the hardest and slowest claim is investment performance. Start with the operating changes that can be observed sooner.

Insurance triage can move quickly while leakage, reopens, external-adjuster expense and adverse outcomes accumulate somewhere else in the claim.

Straight-through processing can look better while the costly work accumulates where automation hands off, fails or needs a second look.

A predicted yield gain is a hypothesis. The value case travels through the recommendation, batch, quality outcome, energy, scrap and financial record.

In grocery, forecast accuracy becomes a business claim only when it changes a replenishment or labour decision and survives the economics of shrink, availability and markdowns.

Once AI changes real work, a project list is too small a unit of management.

Not intelligence — whether it can alter work, spend or a customer record.

A model generates. A harness supplies context, state, tools, permissions, approvals, recovery and the record of what happened.