Field note · Fractional CTO leadership

Paying down technical debt without a rewrite: what works when the business cannot stop

Almost every company I join has a system everyone wants to rewrite. Almost none of those rewrites should happen. Debt is real and expensive, but the cure is usually a steady, funded paydown that the business can see, not a big-bang replacement that freezes product work for a year. This is how I approach it, and how I explain it to founders and boards.

Field noteSumanta Mukhopadhyay8 min read6 sources
#leadership#architecture#platform
FIELD NOTE · FIELD NOTE · FRACTIONAL CTO LEADERSHIP17 Nov 2026 · 21:20 GST

The rewrite that never ships

The pattern is familiar. A core system has grown slow to change and fragile to run. Engineers propose a rewrite. Leadership agrees, because it sounds decisive. A year later the new system handles some of the work, the old one still handles the rest, and the team now maintains both. Product work slowed the whole time.

Ward Cunningham, who coined the term, described debt as the cost of shipping with understanding that later needs to be repaid by refactoring; the danger is not borrowing but never repaying1. Repayment works best in small, regular instalments, not one enormous one.

Step 1: name debt by what it costs the business

"The code is messy" is not a reason a board can fund. "Every change to pricing takes three teams and a weekend release" is. I start by listing debt items with their business effect: what they slow down, what they put at risk and what manual work they create. Martin Fowler's distinction between deliberate and inadvertent, prudent and reckless debt helps separate conscious trade-offs worth keeping from accidents worth fixing2.

Risk
Could cause an incident or breachfix first
Speed
Slows important changesfund steadily
Toil
Creates repetitive manual workautomate
Annoyance
Ugly but harmlessleave it
Rank debt by business effect. The bottom layer is where most rewrite enthusiasm lives, and where it should not.

Toil deserves special attention. Google's SRE guidance defines it as manual, repetitive work that scales with the service and has no lasting value5; it is often the cheapest debt to pay down and the easiest to show results for.

Step 2: fund it like a portfolio

Instead of a one-off project, agree an explicit share of engineering capacity for debt and reliability work, and protect it. The exact share depends on the state of the system and the business's appetite; what matters is that it is visible, agreed and not raided every time a deadline looms.

Then measure whether it is working. The four DORA measures, deployment frequency, lead time for changes, change failure rate and time to restore, are a good, widely understood signal of whether paydown is making delivery faster and safer4. If they do not move, change what you are paying down.

Debt paydown that cannot show its effect on delivery or risk will lose its budget. Measure it, report it and connect it to things the business already cares about.

Step 3: replace piece by piece

When part of a system really does need replacing, the strangler fig approach works far better than a rewrite: put a stable interface in front of the old system, build new capability behind it one slice at a time, and gradually route traffic to the new parts until the old system can be retired3.

  1. FrontPut a routing layer or stable API in front of the old system
  2. SlicePick one capability with clear boundaries and real value
  3. BuildBuild it new behind the same interface
  4. ShiftRoute a small share of traffic to it, compare, then increase
  5. RetireRemove the old code path once the new one carries everything
  6. RepeatChoose the next slice based on business value and risk
Each slice delivers value on its own. If priorities change, you stop with something useful, not half a rewrite.

Modern routing makes the shift step routine. Weighted routing can send a small share of traffic to the new implementation, or route specific requests by header, and roll back instantly if something goes wrong6.

Diagram of a load balancer sending requests through an HTTPRoute: normal traffic split 90 percent to service v1 and 10 percent to v2, and requests with a canary header sent entirely to v2
Weighted and header-based routing: the mechanism that lets a new slice take traffic gradually while the old system keeps running. Source: Kubernetes blog, Gateway API authors · CC BY 4.0

When a rewrite is actually right

SituationUsually better
System works but is slow to changeRefactor and replace slices
Platform is unsupported or insecure with no upgrade pathPlanned migration, still in slices where possible
Small system with clear, stable requirementsA rewrite can be reasonable
Business model has changed completelyBuild new alongside, retire old

Even when replacement is justified, I push to run old and new in parallel and move traffic gradually. The goal is never a "big bang" day.

Field note: what I tell founders

Three things. First, debt is a normal cost of moving fast; the question is whether you are paying the interest knowingly. Second, the team that wants a rewrite is usually right about the pain and wrong about the cure. Third, the best signal of progress is not new architecture diagrams but faster, safer releases and fewer late-night incidents. See a fractional CTO's first 90 days, architecture decision records, the legacy estate modernisation case study and the safe change case study.

How I can help

I help leadership teams decide what debt to pay down, fund it sensibly and modernise systems without stopping the business. I am based in India, the United States and the UAE, working globally. See fractional CTO work, how I work, or analyse your challenge and bring the result to a call.

References
  1. 1
  2. 2
    Technical Debt QuadrantMartin Fowler · 2009
  3. 3
    Strangler Fig ApplicationMartin Fowler · 2004
  4. 4
  5. 5
  6. 6
FAQ

Common questions

Should we rewrite our legacy system?

Usually not all at once. Big rewrites tend to take longer than planned, freeze product work and recreate old problems. Replacing a system piece by piece, behind stable interfaces, keeps the business running and shows progress early.

How do I explain technical debt to a board?

In business terms: what it slows down, what it puts at risk and what it costs in people's time. Tie each debt item to a measurable effect, such as release frequency, incident rate or hours of manual work.

How much engineering time should go to technical debt?

There is no universal number. Agree an explicit share of capacity with the business, protect it, and adjust it based on evidence such as delivery speed and incident trends.

Keep reading

Related

ServiceFractional CTOI work as a part-time CTO for founders, CEOs and boards who need experienced technical judgement now, and a team that can carry it on after I step back.→InsightMicrosoft drops older Oracle GPUs: what it means for anyone signing AI capacity dealsMicrosoft is winding down a 2023 arrangement to rent AI computing from Oracle, reportedly because the capacity ran on older Nvidia chips that no longer met its needs. If one of the largest AI buyers can find rented GPUs falling short within a few years, smaller buyers committing to multi-year capacity should look hard at what they are signing. Here is how I think about GPU ageing, contract terms and the architecture that keeps you free to move.→InsightThe Binance hack: what a laptop-to-withdrawal attack teaches every platform that moves moneyOn 29 October, Binance reported what is being described as the largest crypto theft on record, about $6.3 billion. Its early account is familiar to anyone who runs payments: a compromised laptop, access to a service container, and withdrawal requests that the system accepted. The lessons apply to banks, fintechs and marketplaces as much as to crypto exchanges.→InsightReddit goes logged-in: how to protect and monetise valuable content when AI answers the searchOn 29 October, Reddit said it will move towards a primarily logged-in experience, citing falling search referrals and scraping of its content for AI. It is the clearest sign yet that the open, crawlable web is being rebuilt into tiers. Any business whose content, data or listings have value now faces the same design question: who gets what, how, and on what terms?→
Open to advisory, fractional CTO & select builds

Let's talk about what you're building.

The fastest route is a free 30-minute call. Prefer to write? Send a message and I'll reply personally within two business days.

Start a conversation

How would you like to reach me?

Book a 30-min callFastest. Pick any open slot on my calendar.→ Write in GmailOpens a pre-filled draft in your browser.→ Use my email appOutlook, Apple Mail or your default client.→

I reply personally within two business days. Time zones from the Americas and Europe to the Middle East and Asia-Pacific.