Agencies ship the agent. The program around it needs two things an agency is not set up for: every team the workflow crosses agreeing on the trade-offs, and written rules for what the agent decides on its own. I did the first part for years as a Technical Program Manager at eBay, working with directors and engineering managers across roughly a dozen teams. Per hour, an agency is cheaper. Per outcome, count the work nobody at the agency owns.
If an agency is the better call, I will say so.
Before you askEvery engagement starts with the AI Program Diagnostic: $2,500 flat, two weeks, credited in full if you move to the build or a retainer within 30 days. Builds run from $7,500 a workflow with a two-workflow minimum, which is the closer comparison to an agency project fee. The Fractional retainer runs from $12,000 a month for one day a week, up to three days, 6-month minimum, with a 14-day mutual exit from day one.
Use these on me too.
An agency is paid to ship the build. The work below decides whether the build gets used, and it sits between teams instead of inside one.
An AI workflow crosses ops, finance, IT, and the people doing the work. None of them report to each other. At eBay my job was getting roughly a dozen engineering teams across Core Product and Core Technology to agree on trade-offs. GAT runs AI programs the same way, with one map and one owner for the dependencies.
Before anything ships, the plan says which steps an agent runs on its own and which need a human to sign off. Autonomy gets earned on eval results. In the family office composite, wire instructions are never auto-populated and callback verification stays a human step.
Budget, vendor choices, and adoption get decided by directors and VPs. I worked in those rooms at eBay, reporting into a VP of Engineering with directors and engineering managers as peers. I bring the trade-offs to that level and get a decision on the record.
AI Workforce Training trains the exec layer and the builders first, then everyone else, on the workflows that shipped. Agencies leave with the knowledge. Your team keeps it.
Someone already owns the strategy, the eval bar, the vendor calls, and adoption. You need extra hands to ship a defined automation faster than your team can.
A single document pipeline, integration, or agent for a workflow someone has already mapped. Project fee, fixed deliverable, 4 to 8 weeks.
Braintrust or an equivalent is already running. You will wire the agency's output into your reliability bar yourself.
A single test to see whether AI moves one metric. A retainer is the wrong shape for that. An agency fits.
If two or more of these show up, a build-only engagement leaves the program unowned.
An agency can build. Nobody will own budget defense, vendor consolidation, or adoption. That work stays, and it goes unowned.
Ops, finance, and IT each own a piece. An agency takes direction from one sponsor. I get the owners of each piece to agree on the trade-offs before the build, and keep them aligned after.
The agent works and adoption stalled. The agent sped up one step and the handoffs around it stayed. That gets fixed by mapping the workflow and rebuilding it, then measuring against the old number.
Each has its own agency, stack, and eval. Nothing ranks them, and nobody set the rules for what an agent may do without a human. The diagnostic puts them on one workflow map and one kill list, with the approval rules written down.
Agencies leave with the knowledge. I hand over the eval suite and the runbook, and AI Workforce Training gets your people running it without me. No permanent dependency.
An AI automation agency builds and ships agents or workflows for a fixed scope. I map the workflow first, ship the build against a measured baseline, and own the work between teams: getting every owner the workflow crosses to agree, setting the governance rules, running the vendor calls, and training your team. An agency delivers the deliverable. I own the program that keeps shipping after it.
When you already have a senior AI leader in-house who owns the strategy, the eval bar, vendor calls, and adoption, and you need extra hands to ship a specific automation. Project fee, fixed scope, fast turnaround.
When you have no AI leader in-house and you need both halves: someone to map the work, set the baselines, defend the spend, and run the eval bar, and someone to ship the production builds. One person for both. Otherwise you stack two vendors and the program never gets an owner.
Usually not. Governance means deciding which steps an agent runs on its own, which need a human sign-off, and who approves a change. That decision sits with your leadership, outside a fixed-scope build. I write those rules into the plan before the build and hold every agent to them with evals.
AI programs stall between teams, not inside them. As a Technical Program Manager at eBay, I worked with directors and engineering managers across roughly a dozen teams to agree on trade-offs and surface risk early. AI programs need the same work: mapping dependencies across ops, finance, and IT, and getting a decision from the people who own each piece.
Per hour, an agency looks cheaper. My builds are priced per workflow shipped, from $7,500 a workflow with a two-workflow minimum, after a $2,500 diagnostic that is credited in full if you build within 30 days. The Fractional retainer starts at $12,000 a month for one day a week with a 6-month minimum. The comparison changes once you count the work an agency does not do: the map, the baseline, and the ownership after launch.
If an agency is right, I will say so and tell you what to ask it.
Book 30 minutes with EdwinWeighing other options? See the other comparisons.