AI is most useful to me when it speeds up framing, documentation and iteration while people still own judgement, risk and accountability.
I use AI in two connected ways: in client delivery, where Salesforce AI, Einstein and Agentforce are becoming part of the martech stack; and in my own product work, where tools like Claude Code help me prototype, document and iterate faster.
Many project problems start with unclear language. A requirement is too broad, a risk is too vague, or a stakeholder update hides the actual decision needed. AI is useful for producing a first draft that can be challenged.
I do not treat the first draft as correct. I treat it as material to edit. That saves time while still keeping responsibility with the person doing the work.
Documentation often fails because people do not know where to start. AI can propose a structure for a runbook, QA checklist, operating procedure or stakeholder summary. Then the team fills it with real project context.
This is especially helpful in migrations, where knowledge is spread across platforms, people and informal habits.
One of my favourite AI uses is asking for objections: what assumptions are weak, what risks are missing, what stakeholder questions might come later. The answers are not automatically right, but they often reveal a blind spot worth checking.
For project delivery, this matters because hidden assumptions turn into late rework.
Side projects like Keto Scanner help me stay close to how modern AI-assisted development feels in practice. The value is not only the app itself. The value is learning how quickly a person can test an idea when AI helps with implementation, interface options and debugging.
That experience makes AI more concrete. It is easier to advise teams on AI adoption when you have used the tools to ship something real.
AI can suggest, summarize and generate. It cannot own the stakeholder relationship, decide acceptable risk, understand political context or be accountable for the outcome.
That is the line I try to keep clear: AI can accelerate the work, but people still own the decisions.
The best use of AI in delivery is not replacing project discipline. It is making good discipline easier to apply: clearer documents, faster drafts, better questions and shorter feedback loops.