Cisco has given engineer Padmanabhan Krishnamurthy an unlimited AI-token budget to test whether a two-person team can refactor 98% of the company’s legacy code within 24 months. That is an ambitious target, not a completed rewrite, and a useful test of whether AI can make software modernisation less of a slow, costly trudge.
Cisco Watch analysis
What happened
Cisco President and Chief Product Officer Jeetu Patel told Fortune that Krishnamurthy, who joined Cisco through its 2025 acquisition of EzDubs, gave himself four weeks to prove the approach. Patel said he had given the engineer unlimited tokens for the work. PYMNTS reports that the longer-term goal is to refactor 98% of Cisco’s legacy code in two years, with a two-person team.
The report also describes a separate example from Itaú Unibanco, Brazil’s largest bank. The bank says its AI-assisted Batch2Cloud project cut migration time for batch systems by up to 70% and sped delivery by as much as 3.3 times in its initial implementation phase. Itaú says its validation framework checked that business-critical rules survived, and that generated code passed its security and quality pipeline on the first pass.
Why it matters
Legacy code is expensive to understand before anyone can safely change it. Cisco’s target asks whether AI tools and a small team can take on that work at unusual scale. Itaú’s reported results offer a concrete example of AI helping with analysis, code generation and validation, while leaving technical decisions and governance with specialist teams.
The stakes are not just whether the code compiles. In banking, a missed rule could affect payments or calculations, so faster migration only counts as progress if the replacement behaves correctly. The numbers here are company-reported goals and results, not proof that every old system is ready for an AI-assisted makeover.
Our read
Cisco’s experiment is worth watching because it puts a striking number and deadline against a familiar enterprise headache. But “98%” is a target, not a verdict. The useful measure will be what gets safely changed, how much human review the work needs, and whether the savings survive contact with production systems.
What to watch
- Whether Cisco reports progress against the 98% target and 24-month deadline.
- How much of the work is completed by AI tools versus human engineers.
- Whether Cisco shares details of testing, review and safeguards before rewritten code reaches production.
Discussion spark: Would you trust AI-assisted refactoring to handle most of a company’s legacy code if engineers still approve the changes, or should a goal as large as Cisco’s be approached more cautiously?
Sources and evidence
- Source update (2 November 2026, 18:59 UTC)
not affiliated with or endorsed by Cisco