The bottleneck moved: what 131 product and engineering teams told us
Guy Leshno · 17 November 2026
Read the originalAutonomyAI's review of sales calls with 131 software companies found that AI made code cheap, while engineering review of non-engineers' changes became the constraint.
AutonomyAI sells AI tools that let product and design staff change a company's real product. Its marketing lead, Guy Leshno, reported on 17 November on 326 of the company's sales calls with buyers at 131 business software companies. Product leaders described handoffs and being shut out of the codebase, while engineering leaders described review they could no longer trust.
- The team logged 1,103 complaints and counted each theme once per company; four in five companies raised three or more.
- Of the 70 companies that raised trust in incoming code, 30 called it an acute problem.
- Prototypes built in separate AI tools were rebuilt by engineers because they did not use the real codebase.
- The report advises writing down who may open and review a pull request, and counting changes from outside engineering that reach production.
A team can measure AI by the changes from product and design that reach production, not by prototypes or lines produced. This is a vendor's account drawn from its own sales calls.
Write down who may open a pull request, who reviews it and what a change from outside engineering must prove, before adding more AI tools.
Count each quarter how many changes from product and design reached production, instead of counting prototypes or lines of code.
Derived by Working Surface from the article; more in the Playbook. Source line: The unit of progress is a merged change.
20 November 2026: as AI agents make prototypes, design tools and code cheap, the scarce work becomes a person's judgement and a named person answering for the result.