Cisco is putting an ambitious AI-assisted software overhaul to an early test: a two-person team has four weeks to demonstrate whether its approach can help address 98% of Cisco’s legacy code within 24 months. Jeetu Patel’s decision to back Padmanabhan “Paddy” Krishnamurthy was notable not just because of his age, but because the former startup founder was working five levels below Patel in the organization.
2
5
6
How Patel came to back Krishnamurthy
Krishnamurthy joined Cisco in 2025 after the company acquired EzDubs, the real-time translation startup he co-founded. Patel reportedly encountered him during a visit to Cisco’s San Francisco office. Their conversation led to Krishnamurthy proposing a small-team approach to modernizing older code; Patel then gave him broad latitude to try it, including unlimited AI tokens.
4
5
Patel described Krishnamurthy as having been “buried away in the organization doing something unstrategic.” The assignment therefore represents a bet on the founder’s proposal and initiative, rather than a promotion based simply on seniority. Reports also describe Patel as believing younger employees can challenge established ways of working, while acknowledging that putting them alongside colleagues with decades of experience can create friction.
5
18
What the four-week test is meant to show
The initial phase is a feasibility test: can a two-person team use AI to refactor code in a way that could support a much larger modernization effort? The reported long-term ambition is to rewrite or address about 98% of Cisco’s legacy code over 24 months.
4
6
That distinction matters. The 98% figure is a target, not evidence that the code has already been rewritten—or that the four-week test has succeeded. The available reporting does not spell out the trial’s precise code scope or acceptance criteria. It establishes a proposed test and an ambitious goal, not a completed migration.
6
7
How the project fits Patel’s AI strategy
The experiment fits Patel’s broader view of AI as more than a productivity aid. In a Financial Times interview, he said he sees AI as a way to “make the impossible happen.” He also described a plan for more AI-written products: one product had been written entirely with AI, with six more expected by the end of 2026 and a goal of 70% of software products being fully AI-written by the end of 2027.
17
Those product targets are separate from Krishnamurthy’s code-refactoring experiment. Together, they show Patel’s appetite for testing ambitious uses of AI, but they do not establish that the two-person team’s approach will scale across Cisco or that the legacy-code target will be met.
What remains unproven
The clearest takeaway is that Cisco is testing an approach, not announcing a finished rewrite. The four-week phase is the first hurdle; the 24-month, 98% target depends on whether that experiment demonstrates a workable path. Until results are reported, the plan is best understood as a high-stakes AI engineering bet—not proof that AI can rapidly replace or modernize nearly all of a large company’s legacy code.