Three Times an Agent Fleet Is the Wrong Answer

Armada Works runs discovery calls most weeks. Thirty minutes, no follow-up sequence, no pressure. The founder (me) talks to the prospect, asks about the bottleneck, and decides whether agents are the right tool. Sometimes they are. Sometimes they are not. And when they are not, I say so on the call rather than collecting a deposit and finding out six weeks later.

This post is the written version of that conversation. Three specific situations where an agent fleet is the wrong answer, explained the way I would explain them across a table.

You Don't Have a Bottleneck Yet

The first thing I ask on every discovery call: what is the work that is not getting done? Not "what would be nice to have." Not "what does your competitor do that you don't." The question is narrower than that. What specific function, right now, is either not happening at all or happening so slowly that it is visibly holding back the business?

If the answer takes more than ten seconds, or if the answer is "well, everything sort of," that is the signal. An agent fleet solves a specific operational bottleneck. Content that is not getting written. SEO that no one is tracking. Outbound that the founder keeps meaning to start but never does. The fleet needs a clear target, because each agent is scoped to a single function with a defined cadence and a measurable output.

When a founder says "I just want to use more AI," that is a different problem. It is a tooling exploration problem, not an operational bottleneck. The right answer there is to spend a week with Claude Code or Cursor, build some personal workflows, and see what sticks. That costs nothing and teaches you more about your own work patterns than any consultant can. If, after that, you find a function that is genuinely stuck, call us. But calling us before you have the bottleneck identified is like hiring a plumber before you know which pipe is leaking.

Your Team Won't Read the Diffs

Every agent in an Armada fleet commits its work to main. Blog drafts, SEO meta-tag rewrites, outbound research, daily briefs. All of it lands as git commits in the client's own repository. The client's team reviews those commits the same way they review any other PR: they read the diff, they approve or request changes, they merge.

This is the mechanism that makes the system trustworthy. It is also the mechanism that filters out companies where agent fleets will not work.

If your team does not have git discipline, if PRs sit unreviewed for days, if "reading the diff" is something the engineering lead does but no one else touches, then the fleet's output will pile up unread. The CMO agent will write its daily synthesis. Nobody will read it. The Content agent will ship a blog post. Nobody will catch the paragraph that drifts off-brand. The Outbound agent will draft a first-touch email. Nobody will notice that the prospect was already in conversation through a different channel.

The fleet produces a volume of work that requires a minimum of human oversight: roughly fifteen minutes a day reading the CMO's morning brief and scanning any diffs that look unusual. That is not a lot. But if the team's culture is "we'll review it next sprint," the system decays. Agents do not get better with neglect. They get worse, because the feedback loop that corrects their output stops functioning.

I have turned down conversations on this basis. Not because the prospect's business was wrong for agents, but because the team's review habits would turn a useful system into a liability. The honest recommendation in that case: fix the review culture first, then come back. The fleet will still be here.

You Want a Contractor, Not a System

Some founders want to hand off a function and never think about it again. "Just handle my content. Send me a report once a month." That is a reasonable thing to want. It is what agencies and fractional hires are built for. It is not what Armada Works does.

An agent fleet is a system that lives in your codebase. You own the repo. You own the agents. You own the dashboard. When the engagement ends, the system stays and you run it yourself. That is the explicit goal: "if we're still here in twelve months, one of us has failed."

This means the founder (or whoever is the system's operator) has to develop a working relationship with the fleet. Not "manage it" in the traditional sense. More like: understand what each agent does, know where the state files live, read the daily synthesis, and occasionally adjust a prompt when the output drifts. It takes less time than managing a junior hire. But it takes more attention than forwarding a brief to an agency and checking back in four weeks.

If you genuinely want to delegate a function and forget about it, a good agency is the right call. They absorb the management overhead so you don't have to. The tradeoff is that you never own the system, the knowledge stays with the agency, and you start from scratch if you switch providers. That tradeoff is fine for some companies. It is just not the tradeoff Armada offers.

Why We Turn People Away

There is a business reason and an honest reason, and they happen to be the same.

The business reason: Armada is pre-revenue, and every engagement has to work visibly well. A fleet deployed into a company without a clear bottleneck, without review discipline, or without any interest in system ownership will not produce visible results. It will produce a folder full of unread markdown. That is worse than doing nothing, because it looks like "AI agents didn't work for us," when in fact it was the fit that was wrong, not the tool.

The honest reason: I have been the founder on the other side of a sales call where the vendor should have said no and didn't. The engagement dragged, the output disappointed, and both sides lost months. The discovery call exists so that neither of us ends up there. If the fit is wrong, I would rather say so in thirty minutes than discover it in week six.

When the Fit Is Right

The companies where an agent fleet works well share a few traits. They have a function that is clearly stuck. They have a team (even a team of one) that reads diffs and gives feedback. And they want to own the system at the end, not rent access to someone else's.

If that sounds like your situation, book a discovery call. Thirty minutes, no pitch deck, no follow-up drip. We will talk about the bottleneck and decide together whether agents are the right tool. Sometimes they are. Sometimes the answer is "try this yourself for a month and call us back." Both are good outcomes.

If you want to understand how the engagement actually runs before booking, how we engage walks through the phases: Pilot, Operate, Build, Transfer. Read it first if you prefer to show up informed. Either way, the call is free and the answer is honest.