Thank you for opening up the latest article in 'The Intelligent Commit' newsletter, my thoughts, research and analysis on Artificial Intelligence and how AI is reshaping the software development lifecycle. I hope you find it insightful, subscribe, and engage.

Last week I attended Axiologik's roundtable event on "AI across the SDLC" in Leeds. The discussion was insightful and the event covered many interesting topics.

One tech leader described a situation following the cost of AI code assistant tool usage climbing sharply in May with the end of the subsidy era (in some cases from $200 per month to many thousands). To control AI spend, the organisation was forced to cap their engineers' frontier model token usage.

Developers quickly hit the limits, and came back with a question.

We can't do any development now. What do we do?

Only recently, maybe 6-12 months ago, those engineers will have been coding the old-fashioned way, by hand. Yet faced with a temporary cap, all development work ground to a halt. This isn't an isolated story; I've heard it a few times in the past few weeks.

As an industry have we become so dependent on AI tooling that we've quietly lost the ability to code without it? And if so, how quickly did that happen?

I then thought about my recent usage. How many of my recent code changes could I have implemented myself very easily, and yet reverted to having my agent of choice do the work for me. I thought of one example last week when it took me longer to work with Claude than just doing the very minor two-line code change myself. Lazy? Or just overly dependent? Worst case, have I lost the confidence to do it myself?

What made the conversation even more interesting was a comment made in the room relating the situation back to business continuity planning. We already plan for the loss of core technology, the moment when systems go down and the business reverts to pen and paper to keep functioning. We document it, we rehearse it, we treat it as essential.

Do we need to be applying the same thinking to AI in software engineering? What is our fallback when the tool is unavailable, whether through cost in this example, an outage, or a supplier failure? Do our teams still have the muscle to deliver without it?

The 2025 and 2026 DORA reports found AI adoption among developers has surged to over 90% and over 75% are using dedicated AI code assistants. If AI is now baked that deeply into how we work, knowing it is a single point of failure is risk management.

As I type this article, access to Claude Fable 5 has just been withdrawn following US government pressure (emerging news, I'm sure the news coverage will catch up). This is a real example of external factors that can impact your teams access to modern tools.

You may argue that coding isn't 'business critical' and that the engineering team will survive without the need for a DR plan. How would you cope if the task-in-hand was to resolve a major system outage or remediate a critical security vulnerability and the AI token cap was reached? Can your development team revert to the 'old way' and fix it by hand?

It's worth remembering too that the dependency isn't just about speed. Researchers found that developers using AI assistants wrote less secure code, while feeling more confident that it was secure. We're not just offloading the typing, we may be quietly offloading our judgement as well, while telling ourselves we're safer than we are.

I don't think the answer is to romanticise manual coding or to resist AI. The productivity gains are well-researched and genuine, and they aren't going backwards. But as with embedding any new third-party service into business critical processes, having a technical dependency on a known single point of failure, without a viable fallback, is a risk all engineering leaders should be discussing and managing.

Thank you for reading my latest article. Please share your experiences and engage with the discussion. I am not always right, but I try to be insightful and thought-provoking.

I'm curious what other engineering leaders are seeing.

  • What impact did the recent price increases have on your engineering function? Did you implement usage caps? What did your developers say when you did?
  • Are you treating AI tooling as part of business continuity?
  • Do your engineers still code confidently without AI, or has that skill started to fade?
  • And where do you draw the line between healthy reliance and risky dependence?

Keen to hear your thoughts in the comments or reach out if you'd like to discuss further.