9 min read · Updated August 2026
AI-native versus AI-augmented, as I've actually seen it play out
One rebuilds the workflow around what the tool can do, the other bolts a tool onto a workflow designed before the tool existed. The difference shows up in the numbers within a year.
Key takeaways
- Bolting AI onto an old workflow usually saves time on one step and adds friction moving between steps, netting out to less than it looks like on paper.
- A practice built around what the tool can do rethinks the shape of the whole process, not just the slowest step in it.
- The tell is where the human checkpoints sit: designed in from the start, or wedged in wherever something went wrong once.
- You can move a practice from augmented to built-around gradually, you don't need to rebuild everything at once.
- The economics diverge slowly at first and then noticeably, usually somewhere around the six to twelve month mark.
What I actually mean by the two terms
People use 'AI-augmented' and built-from-the-ground-up-around-AI almost interchangeably, and the difference matters more than the terminology suggests. An augmented practice takes its existing workflow, the one designed years before any of these tools existed, and inserts an AI step somewhere in the middle: drafting gets AI-assisted, or review gets AI-assisted, while everything upstream and downstream stays exactly as it was.
A practice built around what the tool can do starts from a blank page and asks what the workflow would look like if you designed it today, knowing what a computed module can reliably do. Often the answer isn't 'insert AI into step three,' it's 'steps two, three, and four collapse into one step because the tool handles the handoff between them that used to require a human passing a document along.'
The tell that gives it away
I can usually spot which category a practice falls into within the first ten minutes of watching someone actually work, not from what they tell me. In an augmented setup, the human checkpoints sit wherever something went wrong once, historically, before the AI step was added, and nobody's revisited whether that checkpoint still makes sense given what changed upstream. The review step exists because it always existed, not because someone designed it in deliberately for this new workflow.
In a practice built around the tool from the start, the checkpoints sit at the points where a lawyer's judgment is actually irreplaceable, and nowhere else. That's a design decision, made deliberately, rather than an inherited habit. It usually means fewer checkpoints overall, but each one is doing real work rather than being a leftover from a process nobody's questioned in years.
A concrete comparison I've watched
I worked with two small commercial practices doing very similar first-draft contract review work, on a similar time horizon, using broadly comparable tools. One added an AI review-assist step into their existing process: intake stayed the same, drafting stayed the same, and a new AI pass got inserted between drafting and partner review, flagging clauses of concern. It helped. Turnaround improved by roughly a fifth.
The other rebuilt the intake step itself around what the tool could do: instead of a paralegal manually extracting key terms before drafting began, the intake step now pulled structured terms directly, the drafting step consumed that structured data instead of a loosely formatted intake memo, and the review step checked the structured terms against the draft directly rather than re-reading the whole document line by line. Turnaround improved by closer to half, not because any single step got dramatically faster, but because two handoffs disappeared entirely.
The actual lesson
The second practice wasn't using a fundamentally better tool. It was using the same category of tool, applied to a workflow that had been redesigned around it rather than layered on top of the old one. The tool wasn't the differentiator, the shape of the workflow was.
How to move from augmented toward built-around, gradually
1. Pick one workflow, not the whole practice
Don't attempt a full rebuild across every matter type at once. Choose the highest-volume, most standardized workflow you run and treat it as the pilot for redesigning the whole shape, not just inserting a tool into one step.
2. Map the handoffs, not just the steps
List every point where work passes from one person, or one format, to another. Those handoffs are usually where the redesign opportunity is, far more often than the individual steps themselves.
3. Ask which handoffs the tool can eliminate rather than speed up
Speeding up a step is the augmented move. Eliminating a handoff by having the tool produce output in the format the next step needs directly, without a human reformatting it in between, is the built-around move.
4. Redesign the checkpoints deliberately
Don't just carry over your old review points. Decide fresh where a lawyer's judgment is genuinely load-bearing in the new shape of the workflow, and put the checkpoint only there.
When staying augmented is actually the right call
I don't push every client toward a full rebuild, and I'd be skeptical of anyone who does. For a low-volume, highly bespoke area of practice, the cost of redesigning the whole workflow rarely pays back, because there aren't enough repetitions to recoup the redesign effort. Augmented is the right, sufficient answer there, and building further is just spending effort for its own sake.
The redesign is worth it specifically where volume and standardization are both present: enough repetitions that eliminated handoffs compound, and enough consistency across matters that a redesigned shape actually holds up across most of them rather than needing constant exceptions.
Questions
- Can a small practice realistically build itself around AI, or is that only for bigger firms?
- Small practices are actually well positioned for this, because there's less legacy process to unwind and fewer people who need to agree on the redesign. I've seen faster, cleaner transitions at three-lawyer practices than at much larger ones.
- How do I know if my practice is augmented or built around AI right now?
- Look at your checkpoints. If you can explain why each one sits exactly where it does, given how the workflow works today, you're likely built-around. If most of them exist because 'that's where review has always happened,' you're augmented.
- Is built-around always the better choice?
- No. It's the better choice when volume and standardization justify the redesign effort. For genuinely bespoke, low-volume work, augmented is often the correct, permanent answer, not a stopping point on the way to something more.
- Where should I start if I want to move in this direction?
- Start by mapping handoffs in your highest-volume workflow, not by shopping for a new tool. The tool rarely needs to change, the shape of the process around it does.
- How long does a real redesign like this usually take?
- For one workflow, I've seen it done properly in four to six weeks, including the documentation step. The two commercial practices I described above both got there within that window, with very different results depending on how far they let the redesign go.
Want this built for your practice, not just read about it?