The Week in AI Adoption: Measurement Is the Moat, April 16
Ten posts this week drew the same line. The gap between AI leaders and everyone else is now a measurable number, and proof is the only thing that closes it.
The week in one glance
- The AI proof gap got measured this week. Gartner says 28 percent of enterprise AI projects fully deliver ROI, and PwC says 20 percent of companies are capturing 74 percent of all AI value. Boards have started pricing those numbers.
- On the coding-agent side, ROI first shows up in work that was never going to ship, harness permissions are now a board-level security surface, and the 2027 hiring plan on most founders' desks is already wrong.
- Measurement, not spend, is the moat. The companies pulling ahead are the ones who can name a number that survives a board meeting.
Theme of the week
Measurement is the moat. The posts I published this week kept circling the same point from different angles. Only 28 percent of enterprise AI projects deliver full ROI. Only 20 percent of companies are capturing 74 percent of all AI value. Eighty percent of white-collar workers are quietly avoiding AI tools. Coding-agent ROI shows up in work that never would have shipped, not in the PR count your CTO already brags about. Every one of those stats draws the same line, and the companies on the wrong side of it cannot spend their way across. The board conversation this quarter is no longer about whether to adopt AI. It is about whether you have a number you can defend in September when someone asks what the spend actually produced.
What we published
AI adoption this week
Gartner's latest study put a number on the AI ROI problem, and it is already showing up in enterprise procurement reviews and Series C due diligence. I walked through what a defensible ROI ledger looks like for a Series C preparing for a board meeting or a raise.
Eighty percent of white-collar workers are actively avoiding AI tools, and the data shows it is not a skills problem. It is a trust problem about what adoption means for their jobs. More training will not close that gap. A direct answer to the job-security question will.
Anthropic launched Managed Agents on April 10 with consumption-based pricing, and the unit rate is not the fight that matters. I broke down the four clauses sitting next to it that Series B founders should redline before their next renewal.
OutSystems surveyed nearly 1,900 IT leaders. Ninety-six percent are running AI agents, and ninety-four percent are worried about sprawl. The post lays out what breaks when adoption outruns ownership, and what the Series C teams who are getting scale right are doing differently.
Three states passed narrow AI bills this week. The shape of compliance is shifting from broad frameworks to product-specific bans with criminal liability attached. Series A founders whose products touch therapy, minors, algorithmic pricing, or emotional companionship are now inside laws most of them have not read.
A fresh PwC study shows 20 percent of companies are capturing 74 percent of all AI value. I walked through what the Series B teams in that twenty percent actually rebuild between pilot one and pilot two, because running the same play a second time is what kills most programs.
Two surveys published this week put a number on the gap between AI leaders and everyone else, and the answer your board is already drafting is not about spend. It is about how the top quintile measures value, and what they choose not to do when a pilot stalls.
AI coding agents this week
The 3x productivity question every board is asking is real in some places and invisible in others. Coding agents earn their keep on the work that never got done before, not the PR that was already shipping. I walked through how to find the number that will actually hold up in a board meeting.
Two Claude Code CVEs and a 513,000-line source leak this week signaled the real problem. Coding harnesses are configuration surfaces, and config ships with the repo. The post names the three permissions to scope, sandbox, and allowlist before the next disclosure lands.
When one engineer plus a coding agent ships like three, the 2027 hiring plan on most founders' desks is already wrong. I walked through the reset I run with CEOs, including which roles compress, which roles move earlier, and which ones the harness quietly makes more important.
Signals to implications
Signal. Only 28 percent of enterprise AI infrastructure projects fully deliver ROI per Gartner, and the shortfall is already showing up in procurement reviews and investor due diligence.
Implication. Stand up a per-project AI ROI ledger before your next board pack. [Exec]
Source: The 28% Problem
Signal. 80 percent of white-collar workers are quietly avoiding AI tools, and the data shows it is a trust question about job security, not a skills gap.
Implication. Put the job-security question on the agenda of your next town hall, before the next training program, not after. [Exec]
Signal. Anthropic's Managed Agents shipped on April 10 with consumption-based pricing, and the four contract clauses sitting next to the unit price are where Series B renewals will be won or lost.
Implication. Pull every active AI vendor contract and redline indemnity, data retention, rate-limit, and exit terms before the next renewal window. [Founder]
Signal. Coding-agent ROI first shows up in work that was never going to ship, such as migrations, test coverage, and documentation backlogs, not in PR throughput on the existing roadmap.
Implication. Stop measuring coding-agent ROI in velocity. Start tracking delivered work that was not on the roadmap. [Eng Leader]
Signal. Two Claude Code CVEs plus a 513,000-line source leak this week show that coding harnesses are configuration surfaces, and that configuration ships with the repository.
Implication. Scope, sandbox, and allowlist harness permissions at the org level this quarter, not per-engineer. [Exec + Eng]
The contrarian take
Most of what I saw this week framed the AI gap as a spend or tooling problem. I do not buy it. What hardened this week is a proof gap, not a budget gap, and the companies behind the line cannot close it with more licenses or another pilot. The AI proof gap post makes the point plainly. The top 20 percent are not outspending anyone, they are out-measuring them. And where coding-agent ROI shows up first argues the same truth on the engineering side. The number that survives a board meeting is work that was never going to ship otherwise. So pick one number you can defend in September. Write it down this week.
Next week
If this recap was useful, the newsletter delivers it straight to your inbox every Monday. Subscribe here.