AI Coding Tools Shifted Software's Bottleneck from Writing Code to Deciding What to Build

The Bottleneck Moved

When Anthropic told its growth team to hire more product managers rather than more engineers, according to VentureBeat's industry coverage, it was a quiet signal about a structural change that has been building since late 2022.

Claude Code had, by Anthropic's own account, effectively tripled the shipping capacity of its engineering organization. The limiting factor was no longer how fast engineers could write code. It was how clearly teams could define what they wanted built.

That is a different problem and requires different people.

Five Years of Collapse, Compressed

VentureBeat traced the workflow disruption across five distinct phases.

During what the outlet calls the Stack Overflow era, roughly 2014 through late 2022, the engineering loop was predictable: write a ticket, get stuck, search Stack Overflow, escalate to a senior if that failed, ship. New monthly questions on Stack Overflow are down approximately 77% since November 2022, the month ChatGPT launched. VentureBeat frames that not as a failure of the platform but as the death of the workflow it represented.

The next phase ran through 2024. Engineers used ChatGPT in a browser tab alongside VS Code, which was faster but still fundamentally the same single-threaded loop. The leverage existed, but it was local.

Then Cursor and Claude Code moved the model inside the editor and wired it to the full repository. The senior-engineer escalation path, which had been a career development mechanism for decades, largely dissolved. By 2026, according to VentureBeat, a meaningful share of working developers type `claude` as their first command in a fresh terminal.

The Numbers That Matter

The spec-driven era, spanning roughly 2025 into 2026, produced the numbers that should get leadership's attention.

Amazon's Kiro IDE team reportedly compressed feature builds from two weeks to two days using spec-driven workflows. An AWS engineering team described completing an 18-month rearchitecture project, originally scoped for 30 engineers, with 6 people in 76 days. VentureBeat does not provide an independent source for these figures beyond citing industry reporting, so treat them as illustrative rather than audited. But the directional claim is consistent across multiple reported examples.

In April 2026, Anthropic shipped Claude Code Routines: scheduled, persistent agents that run on a cadence, on a webhook, or overnight. VentureBeat's source material was truncated before describing the full implications, but the pattern is clear. The work is increasingly automated at the execution layer. The human job is upstream.

The Fair Counterargument

Skeptics have a legitimate concern worth stating plainly. Productivity claims from AI companies and their enterprise customers are not neutral. Anthropic has a financial interest in convincing the market that its tools produce 3x output. Amazon has an interest in signaling that it can do more with less headcount. Neither the 77% Stack Overflow drop nor the AWS staffing figures have been independently verified in the sources available here.

More substantively, raw shipping speed is not the same as better software. If engineers are now producing three times the code volume, that code still needs to be correct, secure, maintainable, and aligned with what users actually need. Compressing a two-week feature build to two days means less time for code review, security audits, and edge-case thinking. Whether AI-accelerated teams are shipping better products, or just more of them, is not settled.

That concern is real. It does not, however, contradict the structural point Anthropic is making. If the output quality holds and the bottleneck has genuinely moved upstream, then the skill mix a software organization needs has changed regardless.

What Changes for Engineers

For most of the last decade, product judgment was someone else's job. The PM owned the roadmap. The engineer owned the build. VentureBeat describes this division as something both sides treated as physics.

The division still exists on paper. What changed is that an engineer with strong product instincts can now act on them at a speed that compresses or eliminates the coordination overhead that used to make specialization efficient. The engineer who can write a clear spec, reason about user needs, and hand that context to an AI agent closes the loop faster than a two-role handoff ever could.

The engineers who plateau will be the ones who treated implementation as the whole job.

The Open Question

Anthropicʼs decision to prioritize product management hiring is one organization's response to one organization's tooling. The harder question, which VentureBeat raises but does not answer, is whether the broader industry is hiring for this shift or still staffing for the workflow that existed before November 2022. Engineering compensation, job descriptions, and university curricula were all built around a bottleneck that has measurably moved. How long it takes institutions to catch up is genuinely unknown.