How AI Has Changed My Development Practice (And What That Means for Clients)

A Transparent Account

Clients sometimes ask me directly: "How does AI change how you work? Are you faster? Is the work still the same quality? Am I getting less than I paid for because AI is doing some of it?"

These are fair questions and they deserve direct answers. I've thought carefully about how to answer them honestly, because the honest answer is more nuanced than either "yes, AI is doing your work" or "nothing has changed, I still do it all myself."

The reality: AI tools have changed my practice significantly. The nature of the change is not that AI does the work — it's that AI has changed what kinds of work take how long, and what I can accomplish in a given time. Here's what I mean.

What's Genuinely Faster

The honest list of work that takes significantly less time than it did before AI tools:

Boilerplate implementation: API route handlers, database migration files, TypeScript interface definitions, test scaffolding, configuration files — code that's structurally predictable from its context. I generate this rather than type it. This is genuinely faster, not because the work is less careful but because the mechanical part is automated.

Code comprehension on unfamiliar codebases: When I join a project that has existing code, I need to understand what it does before I can work effectively on it. AI-assisted code reading — asking questions about what specific modules do, getting summaries of complex logic — has reduced this onboarding time meaningfully.

Documentation and comments: I document code more thoroughly than I did before AI tools, because the cost of producing documentation is lower. AI drafts; I review and refine. The result is better-documented code with less time investment than producing the same documentation entirely manually.

First drafts of everything: Functions, tests, schema definitions, configuration, specifications, client-facing documentation — AI produces first drafts that I work from rather than starting from blank pages. This doesn't eliminate my judgment — I still make every significant decision. It removes the friction of getting from zero to something to react to.

Research and technical lookup: Questions that previously required me to consult documentation or StackOverflow I now answer conversationally. The quality is high enough for most questions that I can move faster without sacrificing accuracy.

What Hasn't Changed

The honest list of work that AI tools have not made faster or have made only marginally faster:

Architecture and design thinking: Understanding what a system should do, how its components should be structured, where the boundaries should be, what data model makes sense — this work requires contextual judgment that AI tools don't have. I spend roughly the same amount of time on architectural thinking as before.

Client communication and requirements: Understanding what a client needs, translating business requirements into technical specifications, managing expectations, identifying when requirements have gaps or contradictions — this is human relationship work. AI doesn't help with it.

Complex debugging: Novel, complex bugs that require deep understanding of system state and behavior across components are still time-intensive to diagnose. AI helps with familiar bug patterns; it doesn't meaningfully accelerate debugging genuinely novel problems.

Security review: I review all security-sensitive code personally and carefully. AI tools help with pattern recognition in security review, but I don't accept AI-generated security-critical code without thorough review. The stakes are too high for shortcuts.

Testing judgment: Deciding what to test, what edge cases matter, what the correct behavior should be — this requires domain understanding and judgment that's mine, not AI's. AI helps generate the tests once I know what to test.

What's Actually Different: The Time Budget Shifts

Here's the more interesting way to describe the change: the ratio of how I spend time on software projects has shifted, not the total time for a given outcome.

Before AI tools, a typical feature implementation might have been: 20% design and architecture thinking, 60% implementation (writing code), 20% testing and review.

With AI tools, a similar feature might be: 35% design and architecture thinking, 25% implementation (mostly generating then reviewing), 40% testing, review, and refinement.

The total time for the feature might be similar or somewhat less. But the distribution has shifted. More time goes into the phases that require judgment — design and review — and less into the mechanical implementation phase.

The output of this shift: better-designed software. More time in the design phase means more careful thinking about trade-offs before implementation. More time in review means more thorough evaluation of what was built. The quality I deliver has improved alongside the efficiency, which is what you should expect when design and review get more attention.

What This Means for Clients

Let me be direct about the client implications:

You get more thoughtful architecture for the same investment: When implementation is faster, I can spend more time on the design work that determines whether software is maintainable long-term. This is a quality improvement for clients.

You benefit from better test coverage: AI-assisted test generation means I produce more comprehensive tests than was practical to write entirely manually. Your software is better verified.

Iteration is faster: When a client's requirements evolve — which they always do — changes are faster to implement. The change isn't "free" but the cost is lower, which means less friction in responding to new information.

The thinking and judgment work is still mine: AI tools have not changed what decisions I make about your software. They've changed the cost of implementing those decisions. The architecture, the data model, the security design, the performance considerations — those remain the product of my judgment and experience.

I want to be direct about one more thing: I don't charge the same for a day of AI-assisted work as a day of fully manual work if the AI-assisted day produces less. The fair exchange is: you pay for the value delivered and the expertise applied, not for the number of keystrokes. AI tools don't reduce the expertise required — they change how that expertise is applied.

Why This Makes a Difference for Your Project

The software I build for clients is more maintainable, better documented, more thoroughly tested, and more carefully designed than work I produced before these tools became available in their current form. That's not a claim I make lightly — it reflects the concrete shift in how I allocate time within a project.

For businesses evaluating whether to work with me: I'm not a cheaper developer because AI does work I'd otherwise do. I'm a better developer who can do more, deliver faster, and apply more thought to the decisions that determine long-term software quality. The right question isn't "how much does AI reduce your cost" but "what quality can I get for my budget."

That's the conversation I have with every client. What are you trying to build, what does it need to be able to do, how do we design it to serve you well for the next five years rather than just shipping something functional today?

If that sounds like the developer relationship you want, let's start the conversation at Calendly. I'll give you an honest assessment of what your project needs and what it would take to build it right.