Two of us write this newsletter, alternating Sundays. Every now and then, the shape is call and response: one week someone lays down a theme, the following week the other tries to answer. When that happens, most weeks the response is modest, because most weeks the news does not hand you a gift. This week was generous. Salesforce and Anthropic, forty-eight hours apart, shipped announcements that could have been written inside last week's essay.
Last week we called enterprise complexity what it is. Dried spit with a budget. The residue of humans trying to coordinate through systems too dumb to hold a clean shape. We pointed at workflows, integrations, approval theatre, reconciliation teams. That was the obvious list.
The less obvious item on the same list is your hiring function. This week's news exposed a problem that was already there.
For five years now, production has been getting cheap in waves. Copilot and then Cursor and then Claude Code made code cheap to generate. Deep research products made information synthesis cheap. Gamma and its cousins have made decks cheap to draft. Natural-language queries made ad-hoc analytics cheap. And in two announcements forty-eight hours apart, Salesforce said the same thing about enterprise operations and Anthropic said it about design.
Both announcements are vendor plays as much as concessions. Headless 360 is arguably Salesforce locking in its place in the agentic stack as much as it is admitting the UI was scaffolding, and Claude Design's onboarding quietly absorbs your design system into Anthropic's. Those plays are real. The common admission underneath them is also real.
Each wave hits a different category of knowledge work. Each wave makes the production side of that work meaningfully cheaper to run. The selection side, the function inside every company that decides who gets hired to do the work, has not caught up to any of the waves so far.
An executive somewhere, with feeling, says some version of production is cheap now, we will hire for taste and judgement. "Hire for taste" has been said so many times in the last year it has become a tell. When a sentence is that universal, it has stopped being strategy and become signalling. The interview loop is the honest signal of what the same executive actually believes.
The awkward part is that the research on hiring well is decades old. Meta-analyses going back to Schmidt and Hunter keep arriving at the same finding. Unstructured interviews, which still run most senior hires at most companies, are among the weakest predictors of job performance in common use. Resume experience is worse. Structured, scenario-based interviews work measurably better. Almost nobody runs them. We built Vina, our AI-assisted hiring agent around that gap two years ago, and the gap is still there. What the latest waves have done is make ignoring it expensive.
Here’s the structural rub. Production is testable in an hour. Judgment is testable over months. Every piece of the modern hiring stack is built for the hour. Portfolio reviews reward the candidate who can produce a polished artifact. Take-home assignments reward the candidate who can ship. Coding panels reward the candidate who solves a contained problem under time pressure. The ATS scores for keywords that prove the candidate did things. Whether those were the right things to do is invisible to the scoring.
None of that machinery will change quickly. ATS vendors sell to HR leaders promoted on cost-per-hire and time-to-fill. Recruiters are comped on closes. Hiring managers are measured by how fast they staff their requisitions. The whole structure optimises for throughput. Throughput is the thing that just got automated.
What will actually happen over the next eighteen months is that companies will keep hiring for production capacity, because that is what their pipeline rewards. They will get frustrated when those hires cannot tell them what to build. They will blame the hires. They will do it again. A smaller number of companies will notice the mismatch and start redesigning interviews. A smaller number still will succeed, because redesigning an interview is cheaper than redesigning an HR function with fifteen years of hardened habit around it.
Hiring is just one example of the thing. Performance review cycles are another. Career ladders. Job architectures. The quiet machinery by which companies decide who does what and for how much. Every one of them was built when production was the scarce thing. Every one of them is still running on that assumption. Every wave makes the assumption less true. Headless 360 and Claude Design are this week's two loudest examples. The list is getting longer.
Last week we said the ant colony is not destiny. Neither is any of the playbooks that built it. The hard part of the AI transition is this quiet inventory of management systems that were built for a different set of constraints and are still running on inertia.