Most enterprise AI initiatives stall because executives treat compute like an operational bottleneck, capping budgets before seeing real results. Cisco took the opposite approach when President Jeetu Patel handed 27-year-old ex-founder Padmanabhan Krishnamurthy unlimited AI tokens to tackle legacy code refactoring with AI. Instead of burying him in corporate red tape, Patel gave him four weeks to prove his model and 24 months to rewrite 98% of the company’s legacy codebase.
If you want measurable returns from AI, strict per-seat licenses and tight token caps are holding your best people back. In this article, you will see why granting unconstrained AI compute to high-initiative internal talent yields dramatically higher ROI than standard software rollouts, along with the operational playbook to run this model inside your organization.
Enterprise Technical Debt is Stifling Speed and Trapping Top Talent
Legacy software architectures cripple enterprise speed. Decades of accumulated technical debt force senior engineers to spend workweeks patching fragile code bases instead of building strategic solutions. Standard refactoring roadmaps routinely span three to five years, consuming massive vendor budgets while operational progress grinds to a halt.
The deeper casualty is human capital. High-initiative talent routinely gets stranded deep inside corporate hierarchies, locked into routine maintenance tasks. When Cisco acquired translation startup EzDubs in 2025, former founders were placed five levels down, tasked with low-visibility projects until they proposed running legacy code refactoring with AI using a lean two-person team.
Capping compute resources and enforcing rigid, multi-year roadmaps keeps high-performing builders trapped in defensive work. Enterprise leaders must remove administrative friction and let internal talent fix root operational bottlenecks fast.
Inside Cisco’s 24-Month Plan to Rewrite 98% of Legacy Code
Unblocking talent five levels deep in the org chart
Large enterprise acquisitions routinely absorb startup founders, only to misplace the talent that made the startup valuable in the first place. When Cisco acquired the real-time translation startup EzDubs, cofounder Padmanabhan Krishnamurthy ended up assigned to routine tasks, trapped five levels below executive leadership.
That dynamic shifted when President Jeetu Patel dropped into Cisco’s San Francisco office for an unscripted conversation. High-value technical talent rarely sits at the top of corporate reporting structures. Executives who want rapid enterprise modernization must actively look past internal hierarchies and connect directly with high-initiative builders.
“I’ve given him unlimited tokens because he’s going to refactor code,” Patel told Fortune at a small roundtable dinner. “So I’m like, you can go, you can go nuts.”
Granting direct line-of-sight to the C-suite bypasses traditional corporate friction. When leaders give frontline technical talent explicit backing, software modernizations shift from multi-year committee discussions into immediate execution.
The economics of two-person teams paired with unlimited compute
Traditional technical debt remediation relies on massive engineering teams billing by the hour. Cisco inverted this model by assigning a lean two-person unit to tackle the codebase with unconstrained AI processing power. This shift transfers project expenditure from bloated payroll headcount to raw computational throughput.
Removing strict token limits eliminates approval delays and budget micro-management. Small technical teams can execute thousands of code refactoring iterations per day when they are not waiting on per-seat software licenses or API spend caps.
| Operational Approach | Resource Allocation | Team Structure | Target Timeline |
|---|---|---|---|
| Standard Enterprise Vendor Rollout | Capped per-seat licenses and strict API budgets | Large multi-tier engineering teams | 3 to 5 years |
| AI-Driven Refactoring Model | Unconstrained AI compute spend | 2-person focused execution unit | 24 months (98% of codebase) |
The economic return of legacy code refactoring with AI comes from substituting manual developer hours with continuous automated processing. Investing directly in unconstrained compute for focused internal teams yields faster operational results than adding third-party contractor headcount.
How Operations Leaders Can Deploy the Unlimited Token Model
Auditing unstrategic maintenance workloads
Execs must pinpoint where high-capacity talent is stuck maintaining legacy workflows instead of building competitive advantages. In industrial operations and quality management, this technical drag usually manifests as manual compliance tracking, static ERP data entry, or maintaining fragile database connectors. Identifying these operational bottlenecks requires auditing daily tasks rather than relying on high-level job descriptions.
Jeetu Patel recognized this pattern at Cisco when he found a former startup founder holding a role that kept him tucked deep within the enterprise. The executive realized that valuable technical talent was spending time on low-impact tasks while critical infrastructure stalled.
“I’ve given him unlimited tokens because he’s going to refactor code,” Patel told Fortune. “So I’m like, you can go, you can go nuts.”
Plant directors and operations executives face this exact challenge when skilled engineers spend thirty hours a week patching old systems or processing routine logs. Capping AI access with restrictive monthly per-seat budgets guarantees those high-value employees remain trapped in repetitive maintenance loops.
Standard enterprise deployments treat AI tools like traditional SaaS seats, capping spending at fixed monthly limits per employee. This model fails during legacy code refactoring with AI because modern LLMs consume massive context windows when reading legacy codebases, parsing dependencies, and rewriting spaghetti logic. A single deep refactoring session can burn through a standard monthly API quota in three hours. When an engineer hits a token cap, their workflow stops cold, and they return to manual line-by-line debugging.
The economics of removing these caps are straightforward. High-initiative engineers often cost companies upwards of one hundred dollars an hour. Spending fifty dollars in API usage in a single afternoon to automate three weeks of legacy database migration is an immediate net gain. Enterprise finance teams often obsess over preventing runaway token usage, yet they happily absorb the hidden cost of a senior developer spending forty hours manually updating obsolete Python scripts or fragile database queries.
Removing token ceilings changes how high-initiative talent operates. When developers know they have unlimited compute, they build autonomous refactoring pipelines, feed entire system architectures into context windows, and run thousands of automated unit tests against legacy endpoints. They move from treating AI as a simple autocompleter to using it as an automated engineering engine. The return on investment does not come from distributing tiny AI access to thousands of passive users. It comes from giving an uncapped firehose of compute to the small subset of your team who actively want to tear down technical debt.
Ready to find AI opportunities in your business?
Book a Free AI Opportunity Audit. It is a 30-minute call where we map the highest-value automations in your operation.
The Strategic Shift from Headcount Budgeting to AI Compute Budgets
Why capping token usage limits business upside
Traditional enterprise procurement treats software as a predictable cost per headcount. Executives routinely evaluate digital tools through fixed monthly seat licenses, projecting expenses strictly based on team size. Applying this rigid framework to advanced generative models starves high-capacity teams of the raw compute power required to automate heavy operational workloads.
When leadership imposes strict token budgets, technical teams alter their daily behavior to conserve allocations. Engineers limit model iterations, skip secondary automated validation passes, and revert to manual line-by-line troubleshooting when monthly token thresholds near depletion. Artificial cost caps actively convert an exponential efficiency driver into a rationed, low-velocity utility.
| Budgeting Model | Primary Metric | Operational Impact |
|---|---|---|
| Per-Seat Licensing | Cost per employee | Predictable expense, capped output speed |
| Unconstrained Compute | Throughput per hour | Variable expense, accelerated execution |
Treating compute as a variable performance engine rather than a fixed operational expense fundamentally changes business output. Removing compute restrictions encourages high-initiative workers to run deep, recursive model prompts that eliminate systemic technical debt in days rather than quarters.
Measuring ROI through cycle-time collapse
Evaluating AI investments through software expense reductions misses the true economic opportunity. The actual value lies in radical cycle-time collapse across engineering, quality management, and plant operations. When legacy code refactoring with AI accelerates modernization timelines, saved engineering hours and faster time-to-market far outweigh raw token processing costs.
Cisco recognized this structural shift by changing how executive leadership resources internal talent. Rather than requiring individual project approvals for API consumption, they removed financial friction entirely to achieve maximum execution speed.
“So I’m like, you can go, you can go nuts.”
This decision reframes enterprise software procurement. When leaders evaluate AI performance through throughput metrics, such as legacy modules retired, compliance checks completed, or pull requests merged, compute expenses become minor operational details compared to gains in speed.
Operations leaders must move past conservative per-seat purchasing models. Aligning unconstrained compute budgets directly with aggressive operational targets removes artificial friction, allowing internal builders to execute massive transformation projects on compressed timelines.
Source: fortune.com
