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.

We've all been there, haven't we? Standing in a long queue at airport security, nervously checking our watches, wondering if we'll make our flight. The boarding pass check took a few minutes, but now you're stuck waiting for a single, overwhelmed bag scanner to process everyone.

Now imagine the airport proudly announcing they've solved the problem by making the boarding pass check instant. Impressive, right?

Except you're still waiting the same amount of time. All they've done is move the queue from one station to another. The bag scanner was always the real bottleneck. Speeding up boarding pass checks didn't get anyone to their gate any faster; it just made the bag scanner queue longer.

Sound familiar? This is exactly what's happening with AI Code Assistants.

This isn't a paradox, this is a diagnostic.

I'm sure you've all seen the headlines. Study after study is reporting how developers using AI Code Assistants such as Cursor, Claude Code, and GitHub Copilot are completing coding tasks 40-55% faster. Impressive numbers. And for many developers, it has transformed their daily work, and the latest 2025 DORA "State of AI-assisted Software Development" report found that 80% of technology professionals believe AI has increased their productivity.

And yet, how is the reality in your organisation? Are you flying? Or does it feel… messier?

While individuals report feeling more effective, and quantitative data supports their individual productivity has increased, many organisations report they're not shipping proportionally more value or any faster to customers.

DORA's research confirms this tension. While AI adoption is now associated with an increase in output, organisational benefits remain unclear. Stability has decreased, and initial gains are being wiped out by a significant increase in the time taken through code review.

You'll find articles framing this as the 'AI Code Assistant Paradox'. But here's the tldr.

AI code assistants are not failing you; they're merely exposing what was already broken. This isn't a paradox. It's a diagnostic.

This is Eliyahu Goldratt's Theory of Constraints in action. Improving any non-constraint activity doesn't improve the system, it just creates a larger queue at the real bottleneck.

Applied to software delivery, our primary constraint might be code review capacity, the generation of clear, deliverable requirements, QA bandwidth, deployment pipeline speed, or any number of bottlenecks.

But it's probably not the raw speed of typing code.

AI is a diagnostic tool.

When you give your engineering teams access to AI Code Assistants, you're basically running an experiment on your delivery system, and it may just reveal where your true constraints live.

Scenario 1. Your team has strong foundational practices.

  • Clear requirements and acceptance criteria arrive consistently.
  • Code reviews happen within hours, not days.
  • Automated testing catches issues immediately.
  • Deployment is low-risk and frequent.

In this environment, AI Code Assistants do exactly what the studies promise. Developers code faster and teams ship value to customers faster. Cognitive capacity is being freed and likely redirected to design discussions, better test coverage, and thoughtful code reviews. The whole system accelerates, as DORA's data confirms.

High-performing teams with good foundations see amplified benefits from AI adoption.

Scenario 2. Your system has hidden constraints.

  • Developers are pulling tickets faster than Product Managers can define them.
  • Senior engineers are drowning in pull request reviews.
  • Manual QA processes can't keep pace with code changes.
  • Deployment requires coordination, change windows, and crossed fingers.

Here, faster coding creates chaos. Pull requests pile up awaiting review. Everyone complains that testing is taking too long. The queue of features pending to be released to production keeps growing. Developers feel productive individually, but frustrated systemically. This is what the latest DORA report describes.

"AI creates localized pockets of productivity that are often lost to downstream chaos."

Both scenarios use the same AI Coding Assistant; the difference in outcomes lies in the maturity of your delivery system.

Acting on the diagnostic.

If AI Code Assistants enable our teams to code 55% faster, developers complete coding tasks sooner and free cognitive capacity. The data overwhelmingly says AI makes individuals more productive; what matters is whether we can redirect that capacity to address system constraints.

So, the real question leaders need to ask is what do we do with the time it creates?

Many organisations default to filling the freed time with more feature work, maximising individual output. But this accelerates a non-constraint and buries the system in work-in-progress. A better answer would be to redirect that cognitive capacity toward the primary constraint in our delivery system.

So how do you act on this diagnostic? Here's where to start, in order.

  1. Make your workflow visible. You can't optimise what you can't see, so the best place to start is to map your actual path-to-production. Where does work queue up? Value Stream Mapping is your friend. It's a visual way to track how work flows from idea to customer, exposing where delays actually occur. The primary constraint will reveal itself in the data.
  2. Establish explicit WIP limits. Don't let faster coding create unbounded queues downstream. WIP limits act as 'enabling constraints' that force teams to finish work before starting new tasks. This improves flow and highlights the bottlenecks.
  3. Invest in foundational capabilities before scaling AI adoption. The DORA research is unambiguous that AI amplifies your existing system. If the foundations are weak (e.g. unclear requirements, slow reviews, brittle tests, risky deployments) the introduction of AI Code Assistants may make things worse, not better. Build the foundations first, then accelerate.

These recommendations have nothing to do with AI; they're the same core engineering excellence practices that have always mattered. AI just makes it impossible to ignore them.

In summary, if your developers feel more productive but you're not seeing results, don't blame AI, but thank it for the diagnostic, because it's showing you exactly where to focus your improvement efforts. The question to ask is whether your delivery system is ready to absorb the velocity they create.

Thank you for reading. I'd love to get your feedback.

  • Do you recognise either scenario?
  • Have your teams had success using AI Code Assistants?
  • What problems did you uncover, did you reveal your true constraint?
  • Have AI Code Assistants generated bottlenecks elsewhere?
  • How have you navigated the rollout, what tips can you share?

And importantly, are you stuck at the scanners or flying high towards your destination?