Headcount Follows the Workflow
Infraxio is seven people, down from about thirty before AI, and a full business hub now goes in under three weeks instead of four to six months. The roster moved after the work did. Agents took the repeated slices. People kept the judgment.
Before AI, Infraxio was about thirty people. Today it is seven. What got smaller was the carrying: updating a record, chasing a status, rebuilding the story when a sale became delivery, reminding someone an invoice had never gone out. Those jobs sound necessary. They are the price of a workflow that moves only when a person remembers to push it.
Seven people cannot be that memory and still do the judgment the business sells. I run Infraxio from Ponte Vedra, and the operators we build for show the same line on a hiring plan. The open seat is the person who will stand between the tools and keep them honest. That seat means the workflow still depends on a human bridge.
The roster moved after the work did
Drop a model onto a thirty-person process and the machinery stays: the handoffs, the side spreadsheet, the Friday meeting that decides which draft is true. Headcount follows the workflow. A process that needs translators gets them, and the model is called a pilot.
The step changed here before the roster did. A real step has an owner, a record, and a review, inside the system where the work already happens. IFX Hub holds the deal, the delivery, the tasks, and the path to an invoice, history attached. The part that repeats can go to an agent. Left as a thread of messages, it stays a person's afternoon, and that afternoon is a job.
The handoff is where revenue leaks. Inside our company it was also where the extra people lived. Closing it was an operating decision before it was a product decision. A shared record carries the context. A person keeps the calls a record cannot make.
An agent gets a slice, and a person keeps the call
An agent earns a seat when it is assigned the way a person is assigned. That is the point of putting AI on the task board: a scope, an output where someone already looks, and a human review. A side chat asks for a favor. A task asks for a result with a name beside it.
The slices in our week are plain. Move context when a deal is ready for delivery. Draft the client update from the record. Assemble a first pass of a bid section from the solicitation and material we trust. Flag a lead with no next action. Prepare invoice lines from work marked done. That is the afternoon that used to vanish the moment a client needed a person.
People keep the calls. Whether the client fits. Whether a dock exception is a schedule problem or a promise we should refuse. Whether a pursuit is worth writing, and whether the draft missed what the client meant. I still read the work. A person is the last signature.
Hand a department to a model and the output is too large to check, so the department stays. Hand over a slice, with a person on the close, and the slice can leave the roster. Seven people holds in that shape.
Under three weeks is what delivery looks like now
A full business hub used to take four to six months. We deploy one in under three weeks. Those months were translation: a CRM stitched to projects and billing, kickoffs that re-explained the sale, spreadsheets reconciling numbers the systems would not share. Bad data still makes a fast wrong screen. What left the calendar was the bridge, because the bridge is the product. A state change in Hub replaces the export, the retype, and a second telling of the sale.
A team this size cannot rebuild the same job and call it custom. I wrote about that shift from client work to products. Growth7 lands the campaign and the next action in operations, or the lead sits. IFX Bid keeps the pursuit, the section owner, and the write-or-pass decision where the response is assembled. DockOps keeps the appointment, the check-in, and the yard at the door. Custom systems stay custom when the advantage is how specific the operation is. The hub deployment is the repeated pattern. The exceptions are what we still learn in the work.
Revenue edits the roadmap in the same month
The products were funded from revenue and partnerships, never venture capital. Payroll comes from work that is already real. Nothing arrives later to staff the gaps while a workflow gets cleaned up. If a coordinator is still required, the current work pays for that coordinator in the same month.
A roadmap built around a raise can hire the seams and call that scale. Revenue and partnerships removed the delay. We built Hub because we had to live in it, and the growth, bid, and dock systems because those jobs kept coming back. Building for our own business first is the test: if we will not live in it, I will not ask a client to. A partnership puts someone else on the delivery, so the work still has to fit a small team.
How to tell the process can actually shrink
A few checks show whether a team can get smaller and still do the work.
- Name the slices that repeat. The input already lives in the system, a person can check the output, and a miss shows itself. "Help the team move faster" is a mood. "Draft the kickoff from the closed deal and put it in review" is a slice.
- Name the judgment people will keep: fit, exceptions, pricing posture, anything that would embarrass you in front of the client. Skip that and the model becomes another place where the business runs on what one person remembers.
- Put the slice where the work already happens. A separate chat asks someone to leave the job and carry an answer back. The task, the bid, the campaign, and the dock schedule are already open. Operations software has to work where the work happens, and so does an agent. A system people open after the fact is why adoption is the real go-live.
- Run a full cycle with no meeting to reconstruct it. Lead to job, finished work to invoice, pursue or pass recorded before writing starts. If the meeting is still how the truth gets assembled, the meeting is the system. The team is not ready to shrink.
The hire Jacksonville keeps posting
Ponte Vedra is home. From here the work runs through Jacksonville and the rest of St. Johns County — yards, service firms, solicitation teams, shops past what an office manager can hold. Northeast Florida keeps posting the person who carries context: field notes into invoices, a rewrite of what sales heard, the real schedule living with the dispatcher because the office board is behind. Put that hire into the workflow. A record the crew will open, slices clear enough to assign, agents on the slices, people on the calls. Often the role was a bridge, and a bridge can live in the system.
Smaller is what is left
Seven people is what remained after the workflow changed. About thirty fit a company run from inboxes and from whoever had been there long enough to translate them. Under three weeks for a full business hub is delivery with that translation out of the project. Revenue, partnerships, and never venture capital kept the old shape from being funded while we were still describing the new one.
Add agents to a thirty-person process and leave the process alone, and the staffing comes back with the drafts. Unreviewed work piles up in a tool nobody lives in, and the coordinator stays to hold the thread. Assign the slices, including to an agent, and keep a person on the signature. The headcount will tell you whether the workflow actually changed.