Where AI helps the work
Less repetitive work
Routine implementation, research and documentation can move faster, leaving more time for the decisions that shape the result.
Repeatable checking
Tests, page checks and review passes can be run consistently across more of the work, with issues brought back for human judgement.
More room to iterate
Ideas can become working versions sooner, so we can learn from something real before investing in the wrong direction.
Method
Where AI helps in real delivery
I have spent more than ten years working across websites, product marketing and engineering. AI helps me move through research, planning, implementation and quality checks with less repetition, but it does not supply the customer context or make the commercial decisions. I review the work, refine the structure and test the finished experience. See the wider build process.
Tools
Different tools for different jobs
I use Codex for hands-on implementation and longer delivery tasks, Claude for architecture and close review, and Grok in fast or high-reasoning modes when rapid exploration or a second view is useful. MCP servers can connect approved tools and project context so an agent can do real work without receiving unrestricted access. The mix changes as the products change. See how this applies to app development.
Outcome
What it changes for a project
The main benefit is a tighter learning loop. I can explore an approach, turn it into a working version, check it and improve it sooner. That creates more room to test assumptions and polish the important details within a sensible delivery plan. Timelines and pricing still depend on the scope, risk and integrations involved. See how I approach website redesigns.
Responsibility
What still needs a person
Positioning, design taste, business logic, security and trade-offs all depend on context. They also need someone willing to challenge a weak suggestion and take responsibility when the easy answer is wrong. I use AI to extend experienced judgement, not to hide the absence of it. See how this thinking applies to automation.
A faster loop, with clear ownership
- Plan with precisionDefine the audience, product constraints and technical approach before asking a tool to produce anything.
- Build with AI assistanceUse Codex, Claude and other tools where they can accelerate the work without obscuring the decisions.
- Test and polishRun repeatable checks, review the result in context and polish the customer experience before launch.
Tell me what you need to build or improve.
Share the goal, the problem and what you already have. I will suggest a practical next step.