Cisco's president Jeetu Patel has put a former startup founder in charge of rewriting nearly all of Cisco's legacy code and taken the cap off his AI spending. Generating that much code is the cheap part of the plan. Proving it still does what the old code did is where it will succeed or stall.
Cisco's president Jeetu Patel has given Padmanabhan Krishnamurthy, a 27-year-old engineer, unlimited AI tokens and a mandate to rewrite 98% of Cisco's legacy code within 24 months, Fortune reported Friday. Krishnamurthy thinks two people can do it. He has four weeks to show the approach holds up.
Patel told Fortune he lifted the usual limits because the work is refactoring at Cisco's scale. "You can go nuts," he said he told Krishnamurthy.
Patel found him five levels down
Krishnamurthy, who goes by Paddy, joined Cisco when it bought EzDubs, a Y Combinator-backed startup he co-founded that translated speech in real time in the speaker's own voice. Cisco announced the deal last November and moved the team into Webex, according to the Webex blog.
Inside Cisco he ended up five levels below Patel, in a role Patel considered unstrategic, Tom's Hardware reported, citing Fortune. The two met at Cisco's San Francisco office, where Krishnamurthy pitched the two-person rewrite. Patel said nobody on his own team, including its most committed AI users, had been thinking at that scale.
Cisco has been heading here for over a year
At Cisco Live in June 2025, Patel said he wanted Cisco's engineers to write half as much code by hand and pass the routine work to AI agents, The Register reported. He also admitted Cisco didn't have enough engineers to build everything it had planned.
This May, Cisco said it would cut fewer than 4,000 jobs, under 5% of its staff, in the same week it posted record quarterly revenue of $15.8 billion, Ars Technica reported. CFO Mark Patterson told investors the restructuring wasn't about savings, and Cisco tied it to a shift in spending toward AI and security.
Patel presents Krishnamurthy as the kind of hire he wants more of. He told Fortune he's looking for early-career people who push back on how things are done, and that he deliberately creates friction over ideas while keeping it away from personalities. In his view, Cisco's veterans, outside hires and AI-native founders get more done working together than any one of those groups would on its own.
Writing the code is the cheap part
With no cap on tokens, producing new code costs Cisco very little. Checking it will cost a lot more.
Cisco's software runs routers, switches and firewalls inside corporate networks, carriers and data centers. Code that has been in service that long has picked up years of fixes for particular customers and unusual edge cases. A rewrite has to reproduce all of that behavior, and two people will have to show that it does. Readers in the Tom's Hardware comments raised the problem within hours, with one arguing that regression testing alone would push the project well past 24 months.
Coding agents with broad access have also caused real damage at other companies. Tom's Hardware has covered agents that wiped a company's production database and a developer's production setup. At Cisco, a mistake like that could reach customer networks.
Earlier this month we wrote about a Meta engineer who says he now ships code he doesn't fully understand, because AI-assisted work moves too fast to review every line. A two-person team rewriting most of Cisco's codebase will run into the same limit, at much higher stakes.
Our take
At State of AI, we think Patel has the people side of this right and is taking a big risk on the scope.
Most large companies say they want to find talent buried in their org charts, and few actually pull someone up five levels and hand him a problem that matters. Patel did, and other enterprise leaders should take note.
The unlimited tokens are getting more attention than they deserve. Amazon has told its engineers that token counts are a poor measure of AI use, as we covered this week. On a migration like this, the useful constraint is whether the output can be trusted, and a spending cap was never going to answer that.
What we'd want from the four-week test is a narrow target and hard evidence. Pick one product line, rewrite it, and show that it passes the same tests the old code passes, with numbers on defects and time saved. If Cisco can show that, every enterprise sitting on decades of old code will want the playbook. If the proof of concept is judged on how many lines got rewritten, it won't tell anyone much.
What comes next
Fortune didn't report when the four weeks started. Cisco also hasn't said which code goes first or how it will judge the results. Watch whether the team stays at two people once production systems are involved, and whether Cisco says anything publicly when the month is up.
Patel gave Krishnamurthy an open budget and told him to go nuts. A month from now, Cisco should know whether what he builds holds up.
