Inside banks, insurers and universities, most of them within the last year. The work is building where the data already lives, getting it through security review, and handing it to the team that will run it.
Will West
Enterprise AI EnablementApp DeploymentDepartmental EvolutionCustomer Growth
A keynote build, a pipeline audit, and a podcast on governed agents.
Domo put a live, unrehearsed build inside its CEO keynote. I wired five agents to governed enterprise data with Claude Code, on stage with Josh James, fixing and extending a running Domo instance in front of the company's largest audience of the year.
If this is the shape of the problem you have, I'd like to hear about it.
Start a conversationHow engagements start, and why the second one costs less.
Most of the useful work happens before anyone writes code. I spend the first sessions on the environment with their data and security people: the permissions model, where the data sits, and what stalled any previous attempt.
Teams in regulated industries have been burned by something that demoed well and then could not be deployed. That history tells me what they will agree to next. When a deployment has to clear a bank's security review, the work is everything around the model.
From there the build happens in their systems rather than a sandbox. I hit the connectors, authentication, pipelines and RBAC constraints early instead of at the end. I can connect an agent to a customer's live data in five minutes, sometimes ten; a usable prototype takes an afternoon, and production runs one to four weeks.
At that pace people react to working software. They get specific about what they want once they can see their own data moving, and the questions after that first session are the ones worth building around.
Every engagement produces pattern libraries, implementation playbooks and readiness checklists. The next deployment inside that organization starts from those, and moves faster than the first one did.
The running cost compounds the same way: as the reusable components get tighter, token consumption falls and the model tier each task requires comes down with it.
Classification, extraction, routing and retrieval make up most of the volume. They run below frontier capability, which is why local models keep taking more of them. That makes engineering quality and inference spend the same lever, and most companies still budget them separately.
The job is finished when the team doesn't need you.
Build the capability with the people closest to the problem. If a deployment only works while its builder is in the room, the team still depends on one person.
At Domo that meant building the practice, not only the deployments. Inside a year: a repeatable five-phase delivery model, an engineering team of twenty, a forty-seven-person Center of Excellence, and the build standards and hiring bar underneath them.
Those sixty-seven people now run the work; several hundred more have come through the training beneath them, across every department and inside customer organizations.
For the customer the deliverable is a team that can extend it: patterns they can copy, standards they hold each other to, and enough hands-on time that the next workload doesn't need a phone call.
Hackathons work better than documentation. Sitting with a customer's team in Japan beats shipping them a runbook. Capability spread that far outlasts the engagement; a single expert becomes the constraint.
Named accounts, grouped by industry.
Built at Domo and through independent consulting.
I came to engineering through sales.
Discovery, executive rooms and carrying a number are where I started, and why I'm useful in the room before anyone opens an editor.
Grew a 250-account distributor territory 130%, past $1M a year. Learned to read what a buyer needs, and what their team will adopt.
Scaled top accounts past $100K a month; won Account Manager MVP and a SOAR award.
Led nine to thirteen Host Success Managers on the largest account book in the company, coaching the team to run on data.
Led the company-wide migration from Salesforce and Tableau to Domo and built the Redshift business logic underneath it: the seat where selling turned into engineering.
Hired to build: full-stack custom application work and API connectors for enterprise accounts, delivered to spec on fixed-timeline contracts.
Production AI built inside enterprise customer environments, and the forward-deployed practice that delivers it.
Two places the economics have already changed.
AI can do the task. The open question is how much of the operating model you rebuild once it can.
Most core SaaS functions a company pays for every month can now be rebuilt in-house, fitted to how that team works, and owned outright. The build-versus-buy line moved, and most companies have not caught up.
The wins are in the workflows nobody documented: the handoffs, the approvals, the copy-paste between systems. Agents with human-in-the-loop checkpoints take the mechanical parts and give people back the judgement calls.
Fixed scope and a fixed timeline, agreed before the work starts.
I get AI systems from an idea into production inside your environment: agents wired to governed data, custom applications, and the pipelines and permissions underneath them.
Discovery with your data and security people first, something running against real data inside the first week, and production in one to four weeks for most scopes. Remote from Phoenix, on site when the work earns the trip.
A short call about what you are trying to build and what has blocked it so far. If it fits, I scope it in writing before either of us commits.