AI gets a lot of hype. In agency and client work, I have seen it help in some places and fall flat in others. Here is an honest take on where it added value and where it did not - and how that shapes the way I sell AI prototyping versus a full production build.
Where AI helped
Drafting and summarising: First drafts of copy, meeting notes, and summaries of long documents. Not final-word quality, but a real time-saver when you edit and fact-check.
Routine classification: Tagging, categorising and routing content or tickets. When the rules are clear, a small model or script can do the first pass.
Research and ideation: Quick exploration of a topic, competitor angles or "what if" scenarios. Good for sparking ideas; you still need to validate and refine.
Code and scripts: Boilerplate, one-off scripts and repetitive dev tasks. Again, with review - it is a productivity boost, not a replacement for judgement.
Clickable prototypes: Fast explorations of a journey or feature before anyone commits to production. That is the useful version of an AI website redesign: prove the idea, then build properly.
Where it didn't
Client-facing final copy: Tone and brand matter. AI output often sounds generic or off; it needs heavy editing. For anything that goes straight to the client, I still write or heavily revise.
Decisions that need context: Resourcing, prioritisation, "is this done?" - those need human context. AI can suggest, but it does not know your team or your client.
Highly regulated or sensitive work: Where accuracy and accountability are critical, I do not let AI drive the answer. I might use it to draft or research, then I own the outcome.
When a simple rule works: Sometimes the best "automation" is a clear checklist or a few if/then rules. Adding a model is overkill. That is often a process problem, which sits with agency process consulting.
Takeaway
Use AI where it adds leverage: drafting, summarising, classification, research, boilerplate, prototypes. Do not use it where judgement, brand or accountability matter most without human review. And do not assume every problem needs AI - often the fix is process, clarity or a simple script.
If you want to prototype AI in your workflow, I can help: AI prototyping. If the real issue is how pitches and handovers move through the team, start with process, not another tool.
