A common experience with AI agents today is that you tell an agent to go do something for you, and it handles an impressively large portion of the previously manual work. But there’s still some extra work to do to finalize the task, so you go in and review the work, clean up any issues, or adjust the direction that it’s going in, and incorporate the work into a larger work product.
While the AI agent accelerated much of the total mass of work that has to get done, there’s still a last mile required to generate value. In coding, it could be doing the code review, security review, or getting the software into production. In consulting, it's reviewing the research and the powerpoint deck that was generated (and delivering it). In sales, it's doing the customer follow-ups and meetings after an agent has automated the front-end of the outreach process.
The specific ratio doesn’t matter too much, but on the upper end of automation you could see a 80/20 or 90/10 split of what gets automated vs. what you still do manually in many functions. The key, though, is the last mile is actually still incredibly complicated and generally requires domain expertise to understand to the entire task itself, not just the part that’s left over.
This is because this is the part of the work where you have to use judgment and taste to figure out whether the agent’s solution is right, what you should change or edit, or you have to fix things when the agent goes off the rails, and so on. The expertise to oversee this process is the difference between slop and something that’s actual workable and valuable.
A strategic finance analyst can’t just become a CPA because the tool gets them most of the way there, even though they both roll up into the CFO. Equally, it's easy to spot the vibe-coded marketing site where no designer was involved in. The same is true for a larger number of jobs in an organization as well.
Incidentally, this is generally why the AI jobs arguments are wrong. Doing 90% of the work is irrelevant if you haven’t been able to ship the whole thing. The last mile is really what matters.
The new last mile
Ok, so what happens when AI goes from doing 90% of the task to say 95% or 99%? Surely that’s when everything gets automated?
Well, this is an open and debatable question, but my strong hunch will be that we will simply grow the task complexity along with the sophistication of what the tool enables. Today’s 99% solution is tomorrow’s 50% solution.
As we get better at automating larger tasks in the enterprise, this just leads to greater expectations from customers and the market, which then causes an expansion of what the role must do to stay competitive.