From Past Jobs to Better Decisions: How a Remodeler Put Its Experience to Work
The owner once had to connect old estimates, job site decisions, and hard-won lessons from memory. Kernel gave him a company-owned context layer that can draw on that history. Now he can ask questions in plain language and bring what the business has learned into the new project work in front of him.
The Business Ran Through the Owner
Champlain Homeworks is a small remodeling company serving homeowners around Burlington: bathrooms, decks, flooring, repairs, carpentry, tile. The owner runs the business and leads every project, and every project, he said, “is a little bit like starting over from scratch.”
What the company knew about its own jobs lived wherever it had landed. Client decisions in email and text. Materials in online carts, screenshots and messages. Site conditions in photos. Lessons from finished jobs in the owner's head. The decisions had been made. None of it was structured.
The cost showed up in estimating. Prior estimates could be found only by remembering which client they belonged to, and three of five recent projects showed repeated recalculation or option research. Permit rules were worked out again for each town. At the end of a job, what it taught went nowhere. What the company learned rarely survived the job, and the next estimate started without it.
A Business He Can Talk To
Kernel helped Champlain Homeworks establish a durable, company-owned context layer: shared language, job records, operating rules, and lessons from past work that can inform projects underway and work still to come.
The owner currently works with that context through Claude Cowork. He asks questions in plain language, and Cowork uses the company's own material to help him find and apply what the business already knows. To the owner, it feels like talking to an office manager familiar with the company's recorded job history.
The company's context is the durable asset. Claude Cowork is the interface used to work with it today.
What happened last time we worked in this town?
What did we make on this kind of job?
What am I missing?
A change came up.
Owner
Claude Cowork
The owner's actual message to open a new job. Cowork checked the company's folders, assigned the next job number, and asked for the basics in plain words.
Owner: ok i want to add a new client project we are working on
Claude Cowork: Both folders are up to date. This new job will be number 0273. To start its record, I need a few basics: Client name Address or a short name for the job (for example, “Maple Street Kitchen”) Town Type of job (kitchen, bathroom, deck, and so on) Where it stands now: still pricing it, agreed, or work already started? The date it started. This can be the first call or site visit. If you know who's in charge of the job and what it covers, tell me that too. If not, we can fill it in later.
Behind the answers are two things the company owns: a shared vocabulary written in the owner's words, and one record per job that runs from estimate to closeout, where what each job taught gets written down. Asked what he would call it when talking to the field lead, the owner did not reach for Kernel's terminology. He would say check the project info, or look at the job plan.
The system was refined against real work, not hypothetical workflows. As the owner began using it, Kernel adjusted how information was saved and where changes could safely be made. The result fits the way he actually works: conversationally, incrementally, and without another app to maintain.
How Work Feels Now
Before
Now
BeforeRely on memory to connect a new job to past work.
NowAsk what the company already knows about similar work.
BeforeRemember which details usually get missed.
NowHave the relevant checks surfaced during the job.
BeforeKeep lessons in the owner's head.
NowCarry them into the next estimate.
When a new job comes in, the owner asks Claude Cowork what the company already knows about that town and type of work. Cowork brings forward relevant past decisions and flags anything that may be out of date.
Before an estimate goes out, it also prompts easy-to-miss checks: permits, disposal details, and material selections. The owner's principle is simple: “Don't spend judgment time on remembering what we already know.”
His own test for the system is blunt. It fails if it gets stale, or if “there's a beautiful process written down somewhere but everybody is still doing something different in reality.” So updating it is part of doing the work, at closeout, at estimating, and when a change order lands, rather than something saved for a quarterly meeting.
The Outcome
More attention for the house, the client, and the judgment calls in front of him.
The owner is using the system for active work and bringing completed projects into it one job at a time. Each one becomes part of what the company can draw on next time. With the company's history and way of working available through Cowork, he no longer has to hold all of it in his head at once.
Who Owns It
Champlain Homeworks owns the context layer, the records it is built from, and the accounts where both live. The AI provider account is also the company's. Kernel works as an outside collaborator with access the company controls and can remove. At the end of the engagement, that access is removed and Kernel's working copies are deleted.
Claude Cowork is the agent harness the owner uses today, not where the company's knowledge is trapped. If Codex or another compatible tool becomes a better fit, Champlain Homeworks can connect it to the same underlying context. Some tool-specific setup may be needed, but the company does not have to rebuild its vocabulary, operating rules, job history, or lessons from past work.
Company-owned context
Records · Vocabulary · Rules · Lessons
Claude Cowork
today
Codex
later
Another Harness
later
The harness can change. The company's context remains.
Every Kernel engagement starts with the 5-Day Blueprint™: a one-week look at where your company's knowledge lives and what it costs to find.
Start Your 5-Day Blueprint