Your team is using AI like a spellchecker.
Most development teams have tried AI coding tools. They autocomplete a line here, generate a snippet there. Useful, but incremental.
There is another level. One where AI reads your entire codebase, makes changes across dozens of files, runs your tests, and fixes what breaks — while you focus on the decisions that actually matter.
This is what that looks like, from someone who does it every day.
Two levels of AI-assisted development
Most teams are at Level 1. The real gains start at Level 2.
- ✓Completes the line you are typing
- ✓Suggests variable and function names
- ✓Generates a function from a comment
- ✓Answers "how do I do X in Python"
// typing a function...
function validateEmail(email) {
return /^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(email)
}
~30 seconds saved
This is useful. But you are still doing all the thinking. Most teams stop here.
Full breakdown of strengths and limits Coming soon
How I actually work with AI
It is not "write me an app." It is a conversation. I describe what I want in plain English, referencing existing code and design decisions. The AI does the mechanical work. I review every change.
Here is a real example from building this portfolio site:
connor:~/project$
Notice what is happening:
- I told it what I wanted, not how to code it
- I referenced existing components — context matters
- The second prompt was a refinement — like directing a colleague
- I reviewed every change before accepting it
Total time: roughly 5 minutes for what would take 30 to 45 manually. And I stayed focused on design decisions, not syntax.
More examples with full detail Coming soon
What changes day-to-day
Here is what your team's Tuesday afternoon looks like — with and without AI-assisted development.
Traditional workflow
- Morning standup Same as always
- Pick up a ticket Read the requirements, ask questions
- Understand the codebase Grep through files, read docs, trace logic manually
- Write the code Type it all yourself, look up APIs, handle boilerplate
- Write tests Create test files, write assertions, handle edge cases
- Code review Review human-written code
- Debug Read error messages, add console logs, trial and error
AI-assisted workflow
- Morning standup Same as always
- Pick up a ticket Read the requirements, ask questions
- Understand the codebase AI reads the relevant files and explains the architecture
- Write the code Describe what you need, AI drafts it, you review and refine
- Write tests AI generates test cases from your code, you verify coverage
- Code review Review AI output with the same rigour as human code
- Debug AI reads the stack trace, traces the logic, suggests the fix
The meetings are the same. The decisions are the same. The mechanical work — the grepping, the boilerplate, the test scaffolding, the documentation — that is what changes.
Full workflow comparison Coming soon
The honest bit
This is the section the AI tool vendors would rather you did not read. But you need to hear it before you invest.
What AI will not do
- Replace your developers. It makes them faster. It does not make them unnecessary. Someone still needs to know what to build, why, and whether the output is correct.
- Understand your business context. AI does not know your domain, your users, or your constraints unless a developer tells it. It is a tool, not a strategist.
- Make architecture decisions. It can implement a pattern, but choosing the right pattern for your scale, team, and timeline is a human judgement call.
- Guarantee secure code. AI-generated code needs the same security review as human-written code. Sometimes more, because it can be confidently wrong.
What goes wrong without guardrails
- "Vibe coding." Accepting AI output without understanding it. The code works, but nobody on the team knows why — until it breaks.
- Over-reliance. Developers stop thinking critically because the AI "usually gets it right." The 5% it gets wrong can be expensive.
- Security blind spots. AI optimises for functionality, not security. Without human review, vulnerabilities slip through.
Developers who understand fundamentals matter more with AI, not less. The tool amplifies competence — and amplifies gaps.
Full risk assessment and guardrails Coming soon
How to start
You do not need to transform your entire team overnight. Start small, measure, then decide.
Step 1: Pick 2 to 3 developers
Choose developers who are curious and technically strong. Give them Claude Pro subscriptions for a month. Let them use it on real work, not toy projects.
Step 2: Measure what matters
Track tasks completed, time saved on boilerplate, code review feedback. Ask the developers: what was easier? What was harder? What surprised you?
Step 3: Decide with data
After a month, you will have real numbers from your own team, on your own codebase. That is the only evidence that matters. Expand, adjust, or stop — based on what you observed.
The cost framing: Claude Pro is £18 per month per developer. A developer's hour costs significantly more. If the tool saves two hours a week — and in my experience it saves far more — it pays for itself many times over.
At Howden, one of the UK's largest insurance brokers, the development team has been exploring AI-assisted tools — starting with Level 1 autocomplete and seeing real results. The appetite to move to Level 2 is there. The question is not whether these tools work. It is how to adopt them with confidence at scale.
Detailed getting started guide Coming soon
See the difference yourself
Pick a task your team does regularly. See how the time breaks down — and try the prompt yourself.
The proof
Everything on this site was built with AI-assisted development. The portfolio, the game embeds, the interactive components on this page. Browse the projects and see what Level 2 looks like in practice.
See the projectsAbout this page
This page was built using the tools it describes. Claude Code generated roughly 90% of the code and content structure. The hardest 10% — the wording, the framing, the decisions about what matters — that is human.
That is the whole point.