Most firms' first encounter with AI is a member of staff finding a clever prompt. It saves an hour on a first draft, someone shares it in a team chat, and leadership concludes the firm is "doing AI." It isn't - it's doing prompts. Those are not the same thing, and the gap between them is where most of the value in AI adoption actually sits.
A good prompt bolted onto a bad process is still a bad process
If a workflow has an unclear owner, an inconsistent intake step, or a review stage nobody actually does, AI does not fix that - it just produces the same broken output faster. Individual staff finding useful prompts is a sign the firm has appetite, not a sign the firm has a deployed capability. Capability requires someone to look at the whole workflow, decide where AI genuinely removes friction, and redesign the steps around that - not layer a tool on top of steps that were already the problem.
Judgement is the part that doesn't automate
AI can draft, summarise and structure faster than any team member. What it cannot do is decide which output is good enough to send to a client, or which corner of the business is worth redesigning first. That judgement has to sit with someone who understands both the workflow and what "good" looks like for this specific firm - usually a partner or senior operator, not a generic tool rollout.
Where the real return sits
- Not in finding the cleverest prompt, but in deciding which recurring workflow is worth redesigning first
- Not in a firm-wide tool rollout, but in a small number of high-friction processes done properly
- Not in one-off experimentation, but in a workflow that still works the same way in six months, with a named owner
- Not in replacing staff judgement, but in freeing it up for the calls that actually need a person
Firms that treat AI as a procurement decision - buy the tool, hand it to the team, hope for adoption - tend to end up with the same scattered, ungoverned usage a year later. Firms that treat it as a process redesign question end up with fewer, better-run workflows and staff who trust the output because they understand where it came from.
The distinction is simple to state and easy to miss under delivery pressure: prompt experimentation is where AI adoption starts. Process redesign is where it pays off.