How to Be a 10x Engineer in 2027 and Beyond
The author argues that being a 10x engineer means multiplying company impact, not output, and that AI agents mostly produce burnout and low-value code. The real path is spotting projects that were previously too costly or risky and that AI now makes viable.
- Running many AI agents at once likely yields 2–3x throughput, with burnout risk and a flood of low-quality "slop" that can lower company value.
- A 10x engineer is defined by impact: if one change lifts share price by x%, a 10x engineer lifts it by 10x%, so more code alone doesn't count.
- The best opportunities are projects that were previously uneconomical, like an engineer who used AI agents to rebuild a costly licensed component and turned the vendor into a paying partner.
- Historical precedents such as Gmail's JavaScript interface and the Philips iPhone remote show the same pattern: deep knowledge of both the tech and the business, plus spotting a shift early.
The core argument is that becoming a "10x engineer" is about multiplying your impact on the company, not multiplying your output. Sergey Tselovalnikov defines impact loosely: if an engineer lifts the company's share price by some percentage, a 10x engineer lifts it by ten times that. Writing ten times the code only counts if that code moves the business ten times as much. He argues the obvious AI-era shortcut, running swarms of agents to do more work, will likely yield maybe 2–3x throughput, cause burnout, and flood the codebase with low-quality output that can actually reduce company value.
The better path, in his view, is to change what work gets done. That means spotting projects that were previously too expensive, too hard, or impossible, and making them viable. He points to a case Gergely Orosz described on the Pragmatic Engineer podcast: a company paying a large share of revenue to license a critical component. An engineer used AI agents, working from the existing specs and test suite, to build a faster replacement in a fraction of the time a rebuild would have taken. The vendor ended up paying for the new implementation and for keeping it closed-source, which shifted the economics of the whole arrangement. He backs the pattern with pre-AI examples: Paul Buchheit built Gmail's interface in JavaScript despite experienced engineers warning it had always failed, and Philips engineers in 2008 mocked up lamp remote controls as iPhone apps, then realized the phone could replace the physical remote entirely.
The thread tying these together is someone who understands both the technology stack and the business well enough to see a shift in the industry, then act on it before others do. Tselovalnikov argues AI is unusual because it touches nearly every company that builds or uses software, which should give plenty of room for this kind of bet for a while. He also says the core requirements haven't changed: deep knowledge of the business, the technology, and where the industry is heading. The post is a short, opinionated essay rather than a data-driven study, so its claims about returns rest on anecdotes and reasoning rather than measured outcomes.