Cisco gives 27-year-old engineer unlimited AI tokens to rewrite 98% of its legacy code in 24 months

Cisco bets an unlimited AI budget on one young engineer

Cisco has handed a 27-year-old engineer unrestricted access to AI tokens and a two-year mandate to rewrite 98 percent of the company's legacy code, according to a Fortune report relayed by Tom's Hardware. The assignment came directly from Jeetu Patel, Cisco's president and chief product officer, who personally selected Padmanabhan Krishnamurthy — known as Paddy — and, in Fortune's telling, told him to "go nuts".

An acqui-hire pulled from deep in the org chart

Krishnamurthy arrived at Cisco through its acquisition of EzDubs, a translation startup he helped build in 2023. Even with a founder's background, he had ended up five levels down the hierarchy, working on projects Patel saw as far from strategic, according to the report. Instead of leaving that talent in place, Patel pulled him up and gave him a mandate whose scope would have been implausible before the current generation of AI coding tools: refactor nearly all of a major networking vendor's legacy software inside 24 months, with no reported cap on the AI compute involved.

What the unlimited token budget signals

The budget line is the telling detail. Code-generation models are metered by tokens, and heavy enterprise use can add up to real spending — but Cisco has decided that for this project the meter should not be a constraint. That framing treats AI compute the way large platforms treat raw server capacity: inexpensive relative to what it replaces, in this case years of human engineering effort.

Legacy modernization is traditionally among the slowest and most expensive categories of enterprise software work, usually spread across large teams and long timelines. Pointing one engineer with abundant AI capacity at 98 percent of that workload inverts the usual economics of the problem.

The catch: verification, not generation

The target invites skepticism. Decades-old codebases carry undocumented business logic and quiet workarounds, and a rewrite that reaches near-total coverage in two years would somehow have to prove the replacement behaves like the original. In AI-assisted development, code generation is no longer the bottleneck — validation is.

The report, as summarized by Tom's Hardware, does not specify which models are being used, how progress will be measured, how output will be tested, or what becomes of the remaining 2 percent of the codebase. Those details will determine whether the 24-month deadline is a serious engineering plan or an aspiration.

A counter-narrative on young engineers

Tom's Hardware frames the move against a backdrop of AI eroding entry-level roles while youth unemployment climbs. Cisco's bet runs in the opposite direction: rather than treating a young engineer as disposable, the company has given one unusual authority and resources typically reserved for entire divisions. The story, which reached the front page of Hacker News, is likely to be read both as an experiment in AI capability and as a data point on how firms might reorganize around individuals who can operate these tools at scale.

Why it matters

Cisco is effectively running one of the largest real-world tests of AI code generation — not on a pilot or a greenfield app, but on the hardest category of software work there is: legacy refactoring across a giant enterprise. If a single engineer with unlimited tokens can move that needle in 24 months, it resets expectations everywhere for what tiny, well-resourced teams can accomplish. If the effort stalls, it will be evidence that verification and institutional knowledge, not token budgets, are the true constraints on AI-written software. Either way, the two-year clock makes the outcome measurable, and the rest of the industry will be watching.