Claude Code turned all the engineers into three. Now companies need more product thinkers
Anthropic recently told its growth team to hire more product managers, not fewer. The reason, as reported in the industry broadcast, was that Claude Code had quietly turned its engineering organization into a team that shipped about three times its actual value, and the bottle went from the integrated development environment (IDE) to the people who decided what to build.
That detail is easy to miss in the noise of all the AI productivity claims. It is also a structural change in which the entire industry lives. The software bottleneck is no longer typing. Determines what to type. And developers who treat that as someone else’s problem are about to step up.
Many ten years ago, that decision rested with someone else. Software engineering used to be a craft that you used slowly, then practiced in a long, predictable sequence: Go deep into the technology, write the code, ask Stack Overflow if you’re stuck, escalate to a senior engineer when Stack Overflow fails, submit a ticket. The product manager owns the funnel. The developer owns the building. Both sides considered this section to be physics.
Then the funnel folded in five steps.
A brief history of how the engineer’s day was compressed
Stack overflow period (2014 to late 2022): The way engineers thought lived in one place. But new monthly Stack Overflow queries are now down nearly 77% since November 2022, which is no coincidence when ChatGPT was launched. The drop is not a survey on the site. It is a survey about the work flow it represents.
Browser tab period (late 2022 to 2024): The first generation of ChatGPT sat without an IDE. The developers used the same loop they always did, with instant oracle: Write the prompt in the browser, paste the response back into VS Code, repeat. Work was still single-threaded and developer-driven. The average was real but local.
IDE native period (2024 to 2025): Cursor and Claude’s code moved the model inside the editor and gave it access to the full cache. The path to the rise of a great engineer is very fragmented. For years, the prevailing wisdom among veteran developers was that Bash had the longest shelf life of any tool in the stack. In 2026, to get a reasonable share of active developers, the first command entered in the new terminal is claude.
Running season (2025 to 2026): Larger context windows turned one-time work into something that previously required tickets, design documents, and sprints. Amazon’s Kiro IDE team reportedly squeezed a feature build from two weeks to two days using the same spec-driven workflow it submitted. The AWS engineering team described an 18-month redesign, planned for 30 engineers, completed by 6 people in 76 days. The bottleneck is how long it takes to write the code. It was only then that the group could clearly define what the right thing looked like.
Regular times (2026): In April, Anthropic shipped Claude Code Routines: Scheduled, persistent agents that run on a cadence, webhook, or overnight while the laptop is off. Cron returned. The hooks came back. The developer’s job is now part of the orchestration: Spin up a swam before bed, review a bunch of pull requests in the morning. Third-party wrappers like OpenClaw, which was temporarily suspended by Anthropic in April before being partially reinstated, make a similar point on the open source side.
The bottle moved; most groups did not
Engineering has almost tripled. Product management is static. The usual 1:8 ratio of PMs to developers, already abused, is now playing closer to 1:20 efficiency because each developer posts more per day. For example, LinkedIn changed its product manager track with "Product Builder" a program that trains generalists across production, design, and engineering. Anthropic hires more PMs, not fewer. The pattern is consistent across companies that have deployed agent workflows in production: The system generates features that are built faster than making decisions about what to build.
For developers, this is the most important job signal of the decade, and the easiest to miss when productivity news dominates the feed.
First principles are the most important, not the least
The instinct to declare the fundamental obsolete in the age of the agent makes the trend exactly wrong.
If a memory leak slows down production at 3 a.m., and the cause turns out to be a hidden proprietary bug pushed 4 years ago, there’s no agent currently in the field closing that loop at the end. Operating systems, networks, synchronization, and query systems still determine who can resolve an actual incident. They also decide who can see the times when the agent’s output looks good and quiet, expensive, wrong underneath. An agent who has written 70% of the code in a modern repo can’t reliably tell anyone when their assumptions about thread safety, memory ownership, or transaction isolation have deviated from the runtime. An engineer who can read the diff and hold what engineer the whole team needs in the room, and that engineer is built on fundamentals, not on the ability to inform.
What is relevant is that the key is now the ability to use power, not the ability to clean. In 2014, knowing how TCP retransmit worked got the debug ticket closed quickly. In 2026, the same information keeps the entire output pipeline driven by the agent in sending a regression to the scale. The blast radius of an engineer who knows what’s going on underneath goes up, not down.
Reviewing new writing
Developers in 2026 are producing code at a rate that exceeds what any of them can read carefully. A team that ships fast and heavy is a team whose engineers handle AI-generated code reviews at least as well as they ever did writing them. Stack Overflow’s 2025 developer survey put 84% of developers on AI tools, with 46% saying they don’t trust the output, up significantly from 31% last year. That gap, difficult implementation coupled with low trust, is precisely where review capabilities are now most important. Codes that push the lot and update slowly accumulate a debt that will appear during the first real event, and the engineer who can pay it is the one who has paired its volume with a deep knowledge of the first principles of the systems involved.
The new difference is the product funnel
Both of those are necessary. And it’s not enough. The key developer in 2026 is the one who has stopped waiting for the funnel to arrive with a Jira ticket.
That means acting is a role that has historically been allowed to skip.
Talk to customers. Watch how they actually use the product. Read the support line. Stay on the sales phone. The signal that the product team receives from the three layers of the summary, the engineer can still see for himself in the afternoon.
Generate ideas, not just estimates. A product manager who used to get input from 8 developers can’t get input from 20 at the same time. A developer who shows up with a proven, broad opportunity is no longer doing the job of a Prime Minister. The engineer performs the work required by the new measure.
Work backwards from the customer. Amazon has been writing its first press release for two decades. The discipline works well in single-team teams and in multi-agent teams. Both produce a large amount of software that functions improperly without a clear statement to that effect "the customer wins" means before any code is written.
Stop hiding behind bandwidth. Honest answer "Do you have the energy for this idea?" it used to be ‘No.’ With processes, hooks, and a collaborative agent stack, a reliable answer is at hand "What is the importance of opinion?" That’s a different conversation, and it’s very difficult to have without a real perspective on the customer.
What are the rewards for the next decade
The five-stage history above is not really a history of tools. It is a history of what part of the work a person had to do. The human part is, and will always be human for the foreseeable future, up the funnel: From writing, to reviewing, to deciding, to choosing a customer to serve and a problem to solve.
The 2026 version of the senior developer is not a multi-coder. It’s the one that knows what’s built, can prove it’s worth building, and has the agent ships and update command to send it without the system collapsing under its own speed.
The developers who did this internally would spend the next ten years creating the most interesting software ever produced. Engineers waiting for a ticket will spend it watching the ticket being written by the agent next to them.
Ishan Gupta is a software engineer at Amazon.
