Anthropic recently told its growth team to hire more product managers, not fewer, after a quiet but consequential operational shift: Claude Code turned its engineering org into a team that ships at roughly three times its actual headcount. In other words, the throughput engine got faster without automatically upgrading the part of the organization that decides what that throughput should spend itself on.

That is the stake: when software teams can produce built features faster, they do not automatically get better at selecting the right features. The bottleneck moves from the integrated development environment (IDE) to the people deciding what to build. And once that happens, companies that keep treating product judgment as the “slower lane” start collecting a different kind of debt, one that shows up in direction, not velocity.

For most of the last decade, software engineering leaned on a stable division of labor. Engineers owned the build; product management owned the funnel. The workflow was almost physics: engineers dove into technology, wrote code, asked Stack Overflow when stuck, escalated when needed, then shipped the ticket. But the funnels and escalation paths that used to organize work started collapsing in five steps.

First came the Stack Overflow era (2014 to late 2022). Developers solved lots of problems in public, and the site’s answer stream helped define the mental model of how work flowed. Then monthly questions on Stack Overflow fell roughly 77% since November 2022, the same period when ChatGPT launched. The article’s point is not that Stack Overflow suddenly stopped being useful. It is that the workflow it represented got replaced by something faster.

Next was the browser-tab era (late 2022 to 2024), where ChatGPT’s early generation lived outside the IDE. Engineers ran the same loop as before, except the “oracle” sped up: prompt in a browser, paste into VS Code, repeat. This created local leverage, but the process still stayed single-threaded and engineer-driven.

Then the IDE-native era (2024 to 2025) moved the model into the editor. Cursor and Claude Code put assistance inside the repository, and the senior-engineer escalation path largely dissolved. The shift mattered culturally too: Bash had been treated as the tool with the longest shelf life, but the article notes that by 2026, for a meaningful share of working developers, the first command typed in a fresh terminal is “claude.” The spec-driven era (2025 to 2026) made the model even more strategic. Larger context windows turned what used to require tickets, design docs, and sprints into something could be handled within fewer sessions.

The final stage in this evolution is the routines era (2026): in April, Anthropic shipped Claude Code Routines, which are scheduled, persistent agents that run on a cadence, on a webhook, or overnight while the laptop is closed. Cron came back. Hooks came back. And the engineer’s job shifted toward orchestration: spinning up work before bed, reviewing pull requests in the morning. Even the ecosystem noticed. Third-party wrappers like OpenClaw were briefly suspended by Anthropic in April before partial reinstatement, reinforcing the idea that the execution bottleneck had moved, not that automation was “free.”

Here is the operational punchline. If engineering throughput has roughly tripled, but product management “has not budged,” then the effective ratio changes. The article frames it as a traditional 1:8 ratio of PMs to engineers, already strained, playing out closer to an effective 1:20 because each engineer ships more per day. LinkedIn made a similar adjustment by replacing its associate product manager track with a “Product Builder” program meant to train generalists across product, design, and engineering. Anthropic, meanwhile, is hiring more PMs, not fewer. That direction matches what the coverage says is happening across companies deploying agentic workflows: the system produces built features faster than it produces decisions about what should be built.

If you are sitting on a board, running a growth team, or funding the next platform bet, this is where the conversation often goes sideways. Many leaders assume the main risk is “can engineers still do their jobs?” But the deeper risk is whether the human decision layer can keep up with the machine’s output.

That is why the article makes a second, more first-principles argument. In agentic systems, fundamentals are not less important, they are more leverage. When memory leaks take down production at 3 a.m., or when an ownership bug introduced years earlier resurfaces, no agent “in the wild” closes the loop end-to-end. Operating systems, networks, concurrency, and query plans still determine who can resolve real incidents. The engineer who can read the diff and understand the assumptions behind thread safety, memory ownership, and transaction isolation becomes more central, not less.

The corollary is that review gets more important as writing gets easier. Engineers can generate code faster than they can read it carefully. The 2025 Stack Overflow developer survey found 84% of developers use AI tools, while 46% say they do not trust the output, up sharply from 31% the year before. Low trust plus high usage is exactly where review skills matter. The debt is not merely aesthetic. It is the probability of expensive regressions arriving when the first real incident forces teams to pay back ignored gaps.

Finally, the article argues that the differentiator is the product funnel, not just engineering speed. The engineer who matters in 2026 stops waiting for the funnel to arrive as a Jira ticket. That means upstream work historically allowed to be skipped: talking to customers, reading the support queue, sitting in on sales calls, and extracting the signal through three layers of summary that previously took much longer. It also means working backwards from customer wins. Amazon’s practice of writing press releases first for two decades is offered as a discipline that travels well, whether teams are small or using swarms of agents.

So the next decade’s reward is clear in one line: the part that is still human moves up the funnel, from typing to deciding. The honest question leaders can no longer dodge is capacity. With routines, hooks, and cooperative agent stacks, “Do you have capacity?” turns into “What is the idea worth?” And it is a much harder question to answer without a real point of view on the customer.