If you have been slow to automate the parts of your work that require actual expertise, that caution has been well placed. There are two ways to put AI into a business that runs on taste or discernment and the choice compounds in opposite directions.
Both are legitimate, and for a firm whose product is judgment, the second is where the durable advantage sits, because it makes the expert better at the thing clients are paying for.
Deloitte's State of AI in the Enterprise 2026, a survey of 3,235 business and IT leaders, names insufficient worker skills as the biggest barrier to integrating AI into existing workflows. It also reports two-thirds (66%) of organizations seeing productivity and efficiency gains, and 85% of companies expecting to customize agents to fit the needs of their business.
That last figure reports intent. Customization is where a skills gap would bite hardest, because configuring an agent to your business means encoding how your business decides, which makes it a transfer of expertise before it is a software task.
Where each one belongs
Automate the steps where being right is procedural. Formatting, retrieval, extraction, moving a record from one system to another, assembling the first version of a document that follows a known shape. Nobody's expertise is expressed in these, and nobody's judgment improves by doing them.
Augment the steps where being right is a judgment call, meaning the specification where the wrong choice is expensive, the client conversation, and the moment somebody decides which of three directions the work takes. Here the system's job is to put better material in front of the person faster, and then get out of the way.
The failure mode is inverting them: the model takes the judgment work and the person is left approving it. That arrangement erodes the expertise you sell while appearing to save time, and it is what the burnout research in our briefing on what bad adoption costs people is measuring.
Why automation alone is a weak position for a small firm
If the whole of what you do can be automated, so can your competitor's, and the thing you were selling becomes a commodity at whatever the software costs.
Augmentation compounds differently. An expert working with good material produces better work, that work becomes part of the record, and the record makes the next round better. The advantage accrues to your firm, because the material is yours.
The same argument arrives from the context engineering side. A system that makes your people better has to know how your firm works, and the only way it knows that is if somebody in your firm wrote it down. That written record is an asset you own, and it is the one thing a tool vendor can't sell to your competitor.
Why a course won't close the skills gap
Deloitte names skills as the constraint, and the instinctive response is training. Training helps, but it doesn't close this particular gap.
What has to transfer is how your firm decides, expressed clearly enough that a system can act on it. That knowledge exists in the people who do the work, and getting it out of them is a job of documentation and design.
Which is why the practical route for a small firm is usually one real build. You pick a genuine workflow, encode the decisions inside it, and the operating model transfers because somebody had to articulate it to make the thing run.
Related questions
What is the difference between AI augmentation and automation?
Automation removes the person from a step. Augmentation keeps them in it and gives them better material to work with. For firms whose value is expert judgment, augmentation preserves and compounds what clients pay for.
Which should a service business build first?
Automate the procedural steps, meaning formatting, retrieval, extraction and first drafts of documents that follow a known shape. Augment the judgment steps. Inverting that erodes the expertise the firm sells.
What is the biggest barrier to AI integration?
Deloitte's State of AI in the Enterprise 2026 names insufficient worker skills as the biggest barrier to integrating AI into existing workflows.
Can training close the skills gap?
Only partly. The knowledge that has to transfer is how a specific firm decides, which lives in the people doing the work. Getting it into a form a system can act on is documentation and design work.
Does augmentation mean slower adoption?
It means adoption that keeps the expertise inside the firm. Automating a judgment step is faster, and the value moves to whoever supplied the tool.
How we read this
The augmentation case gets made on ethical grounds and we would make it on commercial ones, because the commercial argument is stronger and it survives a skeptical reading.
A firm whose expert judgment is the product has exactly one durable asset, and it is the judgment. Build systems that route around it and you have converted your differentiator into a subscription anybody can buy. Build systems that feed it and the differentiator compounds, because every engagement adds to material only you hold.
This takes longer to feel like progress. Automate a step and you have something visible to point at this week. Augment one and you get a slightly better output that is hard to show anybody until you have a quarter of them behind you.
One caution on the word. Augmentation isn't a reason to keep a person in a step where they add nothing. If somebody's only contribution to a step is moving a file, the file-moving is the automatable part, and keeping them there protects nothing.
What you can do this week
Take one workflow and split it in two columns: steps where being right is procedural, and steps where being right is a personal judgment call.
Then check where you have already put AI. Is it sitting in the second column while a person checks the first? If so, swapping them is usually a configuration change and not a rebuild.
Expect one step to resist the split. There is almost always a step that looks procedural and turns out to carry a judgment nobody wrote down. It is usually the one that goes wrong when somebody new does it. Put that step in your knowledge base before you put it in an automation.
Working together
Flow State Found works with a limited number of businesses to make their best work their baseline. Most firms we speak with arrive after automating part of their work and finding the output looks like everyone else's. We build systems that augment the work instead, letting the perspective you are hired for lead the way.
We take on limited engagements, so it starts with a conversation.
Start a conversation