Approach · Agentic workflows
Let the AI do the reading. Keep a person on the decision.
An agentic workflow hands part of a job to an AI system, such as reading a vendor quote, sorting job photos or drafting a change. We design each one around a bounded task, a named point where a person checks the result, and a hard limit on what the system can do.
How we approach it
- 01
Start with bounded work.
Good first candidates are valuable, repetitive and easy to check. A quote request is one: it matters to the business, and a person still makes every commitment.
- 02
Put the human check where judgment matters.
At D.W. Ross the AI reads a paper vendor quote but never sets a price. A person checks and corrects every field before anything is sent.
- 03
Limit what it can do in the software itself.
That quote tool can create an estimate in QuickBooks and nothing else: no invoices, no payments, nothing sent to a customer.
Questions it answers
- Which parts of this job can software do, and which need a person?
- Where exactly does a person review the result?
- What must the system never be able to do?
- How does the output reach the tools staff already use?
In practice
Where this shows up in our own work, with each item's current status.
- D.W. Ross: photographed vendor quotes read by AI, checked field by field, sent to QuickBooks as estimatesUsed internally
- D.W. Ross: job photos sorted and captioned by AI, approved by a person before they go liveUsed internally
- What a quote request from an AI assistant should carry, and what it must never be able to doPrototype
Tell us what you're trying to accomplish.
We'll talk through the problem, whether Lehvel is a good fit, and a sensible next step. The scope comes from the problem, not a package.