Prasanna S, a software engineer who writes on X, argues that frontier coding models have already crossed a threshold most engineering teams have not registered. In an essay titled “The Great Flattening,” he says the bottleneck in software engineering has moved. It is no longer writing code. It is encoding judgment into the agent harnesses that plan, test, review, and deploy that code.
Two claims sit inside that argument. The first is technical: the scarce skill has shifted from producing code to specifying and supervising the agents that produce it. The second is structural: that shift will compress the org chart, because most of engineering management exists to coordinate the writing, testing, and shipping of code, and multi-agent systems now perform that coordination directly.
His prediction follows from there. As cloud-based multi-agent systems take over the workflows that used to require a full engineering department, the organizations built around those workflows flatten. Layers of engineering management thin out, not because managers stopped being useful, but because the work they coordinated (writing, testing, shipping) increasingly runs itself.
What remains, per Prasanna S, is customer insight and product judgment: knowing what to build, not how to build it. That distinction tracks with what builders have reported anecdotally already, as coding agents take on multi-file refactors and full test suites with less human review at each step. The essay had drawn more than 60,000 views on X within a day of posting, a sign the argument is landing with an engineering audience already unsettled by how fast agent tooling has moved this year.
Flat-org predictions have a history of arriving early. Low-code platforms in the 2010s and offshoring waves in the 2000s promised a similar outcome: fewer engineers, thinner hierarchies, judgment as the last defensible human skill. Neither delivered flattening at the scale forecast. Engineering headcount grew through both waves. It just changed shape and location.
The mechanism Prasanna S describes differs in one respect worth taking seriously. Earlier tooling waves automated execution inside a fixed process; the harnesses he describes are being built to own the process itself, including the review and deployment steps that used to require a human sign-off. That is a claim about scope, not just speed, and it is the part worth testing rather than assuming true because a model can now write good code quickly.
For hiring managers, the practical read is not to hire fewer engineers. It is to hire differently. The essay’s own logic points the premium toward engineers who can specify what an agent harness should optimize for, and away from engineers hired mainly to write code a model can now write faster. Teams built around code-writing throughput alone should audit that assumption before locking in their next headcount plan, not after.
Prasanna S made this argument in an X Article, “The Great Flattening,” published July 14, 2026.
