Louise: I wanted us to have this conversation because everyone has an opinion on AI right now and most of it is either “it will replace everyone” or “it is just a fad.” We think both of those are wrong. We use it every day and the reality is somewhere much more boring and much more useful.
Michael: Boring and useful is a good way to put it. It has not changed what we do. It has changed how long some of it takes.
Daniel: And how much of the tedious stuff I actually get around to doing rather than putting off.
Writing and communication
Louise: This is where I notice it most. We all write proposals, emails, scoping documents. AI does not write them for us but it gets past the blank page. I describe what I want to say, it gives me a structure, and I rewrite it until it sounds like me. That first draft used to take an hour. Now it takes fifteen minutes.
Michael: I use it when I need to explain something technical to someone who is not technical. I know what I want to say but finding the right way to say it used to take effort. AI is surprisingly good at finding that middle ground where you are being accurate without being impenetrable.
Daniel: I use it for the writing I would otherwise never do. Documentation. I know how every system I have built works, but writing it up so the next person can understand it is the job that always gets pushed. Now I describe the system, AI gives me a first draft, I clean it up. The documentation actually gets written. That alone has been worth it.
Louise: Meeting notes too. We all sit in meetings and someone has to take notes. Now we just let AI transcribe and summarise. I get a two-minute summary with action items instead of spending half the meeting scribbling.
Software development
Daniel: Code completion is the obvious one. I use it the way I use autocomplete on my phone. It suggests the next bit and I either accept it or I do not. For boilerplate and repetitive patterns it saves real time. For anything that requires thought, I ignore it.
Debugging is where it has changed things most for me. When something is broken and I cannot see why, I paste the error and the relevant code in and ask what I am missing. Half the time it spots the thing I was staring past. The other half it talks nonsense, but even a wrong answer sometimes points me in a direction I had not considered.
Michael: I have started using it for infrastructure work too. Writing Terraform modules, checking CloudFormation templates for mistakes, generating IAM policies that follow least privilege. The policies it generates are not perfect but they are a better starting point than writing them from scratch. I still review every line.
Daniel: Test writing is another one. I write tests alongside my code and AI generates a decent first pass. It catches the obvious cases and I add the edge cases it misses. The coverage is better than it was before because the boring test cases actually get written now.
Louise: Does it ever get things wrong in ways that are dangerous?
Daniel: Yes. It confidently produces code that looks right, compiles, passes basic checks, and does the wrong thing. That is the main risk. If you are not reviewing what it gives you, you are introducing bugs faster than you were before. AI without code review is worse than no AI at all.
Michael: Same with infrastructure. It will happily generate a security group that is too permissive or an IAM policy with admin access. You have to know what right looks like before you can trust what it gives you.
Running the business
Louise: I think this is the bit people miss because they assume AI is just for developers. We use it across the whole business.
Invoice processing. I used to spend time manually logging receipts and expenses. Now the accounting software pulls the data out automatically. It is not glamorous but it saves hours every month.
Michael: Scoping new work. When a potential client describes what they need, I use AI to help me think through the technical requirements and estimate the work involved. It does not give me the answer but it helps me make sure I have not missed something obvious. Have I thought about authentication? What about data migration? It is like a checklist that asks questions back.
Louise: Email is a big one for all of us. Not sending emails automatically, just drafting responses faster. Most business emails follow the same patterns. AI drafts something, I adjust the tone, and it goes out in two minutes instead of ten. Over a week that adds up.
Daniel: I use it for research. When I need to evaluate a tool or a framework, I start by asking AI for a summary of the trade-offs. It is not always up to date and it sometimes gets details wrong, but it gives me a starting point faster than reading through documentation from scratch. Then I verify the important bits.
Michael: Proposal writing is another one. We all contribute to proposals and the first draft is the hardest part. AI helps structure the response, makes sure we have addressed everything the client asked about, and gives us something to react to rather than starting from nothing.
What it is not good at
Daniel: Architecture. AI can write a function but it cannot design a system. It does not know why you made the decisions you made or what constraints you are working within. I still do all of that the same way I always have.
Michael: It also does not know your business. Your clients, your team, what you can afford to get wrong. It can list options but it cannot tell you which one is right for your situation.
Louise: And it cannot read a room. I can draft an email with AI but I still need to know whether this client needs a phone call instead. That is judgment, not language.
Daniel: The hallucination thing is real too. It makes things up with complete confidence. I have caught it inventing API methods that do not exist. If you do not know enough to spot that, you are in trouble.
Where we think it is going
Michael: More of the same, but better. The tools will get more accurate and handle more complex tasks. But I do not think it changes the fundamental dynamic. You still need to know what you are doing. AI just means you spend less time on the parts that do not require thinking.
Daniel: I worry a bit about the next generation of developers who learn with AI from day one. If you have never debugged something without help, do you actually understand how it works? I am not sure yet. Ask me in five years.
Louise: I am more optimistic about it than either of them. The tools I use now would have seemed impossible three years ago. Whatever comes next will probably make what we are doing now look slow. I just want more of it aimed at people who are not developers.
Michael: If any of this is relevant to what you are working on, get in touch. We are happy to talk it through.