Put An Agent Behind
Everyone On Your Team.
An agent is the part of a system that keeps working when nobody is watching. It takes the repeating half of a job, the part that runs on rules and never gets done on time, so your people spend their hours on the half that needs judgment. We build them into the machine, in your accounts, on your keys.
Most Things Sold As Agents Are Not Agents.
A language model sitting in the middle of a linear automation is still a linear automation. It runs the same play on Friday that it ran on Monday, whether or not Monday worked. That is a useful thing to own. It is not an agent, and it will not compound.
Three things have to be true before we will call something an agent:
- It can see the live business data. Not a spreadsheet somebody exported last quarter. The ad spend, the analytics, the CRM and the payments, in one place, agreeing with each other.
- It decides on a cadence. On its own schedule, not on a trigger somebody has to remember to pull.
- It changes what it does next. The result of the last decision is an input to the next one.
The third one is the hard part, and it is the one almost every build skips. Without it you get something that looks autonomous for a week and quietly gets worse every week after, because it keeps making the same kind of decision long after the market stopped rewarding it.
Easier To Understand As Work Than As Technology.
Every one of these is work your team is squeezing in between the work you actually hired them for, or work nobody has time for and everybody can feel. Read them as capacity and find the ones that describe your week.
-
01 · Attract
Creative That Ships Every Week
TodayCreative goes stale. Your marketer finds time for three new variations a month, the account gets gradually more expensive, and nobody can say which of the old ones is carrying it.
The agentResearches what your market actually complains about, builds variations against it, publishes them, turns off the losers on their own data, and makes more of whatever is working.
Your teamSets the angle and the brand rules, then spends its time on the ideas worth testing instead of the production line.
-
02 · Indoctrinate
One Recording, A Month Of Content
TodayYou record something genuinely good. It sits in a folder. Six weeks later somebody cuts two clips out of it and the rest is never used.
The agentTakes the recording once and produces the clips, the posts, the emails and the article in your voice, then publishes them on the schedule.
Your teamKeeps making the thing only they can make, and stops losing days to cutting it up afterwards.
-
03 · Convert
Every Lead Answered In Minutes
TodayA lead arrives at 9:40pm on a Saturday. Somebody sees it Monday morning. By Monday morning they have booked with whoever answered first.
The agentReplies immediately across SMS, email and DMs, answers the real questions, qualifies against your criteria, books the call, and keeps chasing the ones who go quiet.
Your teamWalks into conversations that are already warm, instead of starting every week by reheating cold ones.
-
04 · Deliver
Every Inbound Read, Scored And Routed
TodayEverything lands in one inbox at the same priority, so the best lead of the month gets the same attention as somebody filling in a form for the third time.
The agentReads each one, scores it against what your closed deals actually looked like, routes it to the right person, and writes the context into the CRM before the call starts.
Your teamOpens the day already knowing who to call first and why, and arrives at the call having read something useful.
-
05 · Ascend
The Ask That Actually Gets Made
TodayNobody asks for the review. Nobody calls the client who went quiet eleven months ago. It is always somebody's job and it is never anybody's priority.
The agentAsks at the moment the client is happiest, follows up on the ones who did not answer, and reopens dormant accounts on a schedule that does not depend on anyone remembering.
Your teamHandles the replies that come back, which was the part that needed a person all along.
-
04 · Deliver
The Monday Numbers, Already Assembled
TodayHours a week pulling numbers out of four dashboards into a slide that nobody reads carefully until something has already gone wrong.
The agentPulls every source into one place, writes the summary, flags what moved and why, and sends it. You can also just ask it questions in plain language.
Your teamSpends the meeting deciding what to do about the numbers rather than agreeing on what they are.
You almost certainly do not need all six. Most businesses have one or two of these quietly eating a whole role, and the rest are fine. Working out which is what the build plan is for.
An Agent Is Only As Good As What It Can See.
This is the part that gets left out of the pitch. An agent making decisions about your ad spend needs the ad data, the analytics, the CRM and the payment data together. Give it one of those and it will confidently optimize toward the wrong number, which is worse than doing nothing, because it looks like progress.
So most of the work is not the agent. It is the plumbing underneath it: getting data out of every tool, into one warehouse, modeled so a lead means the same thing on every screen. That part is unglamorous and it is the majority of the build.
It is also why we do not sell agents as a plugin you drop in. If somebody offers you one that needs nothing from your stack, it is a workflow with a better name on it.
How it fits
Part Of The Build, Not A Separate Purchase.
Agents get scoped in the build plan alongside everything else, priced into the same fixed number, and handed over on the same terms. If none of them are worth building for you, we will say that on the call.
The same handover, applied to agents
- They run in your cloud, on your accounts and your API keys
- You get the code, not a black box you rent access to
- Documentation and a recorded walkthrough, so your team can change what an agent does without calling me
- Every boundary written down: what it can spend, what it can publish, what it must never touch
- Nothing runs on my infrastructure. If we never speak again, they keep running
One honest cost note. Because the accounts are yours, you pay the model and hosting bills directly rather than through us. For most builds that lands in the tens of dollars a month. An agent generating video creative every day costs meaningfully more. You get the number for your specific build before we build it, not after.
The Things People Ask About Agents.
Is this going to replace people on my team?
That is not what this is for, and it is not how it plays out. An agent takes the repeating part of a job: the follow-up nobody gets to on a Saturday, the fifth ad variation, the review request nobody remembers to send. What is left is the part that needed a person in the first place, the judgment and the relationship.
The honest version is that a team with agents behind it absorbs the next round of volume without you hiring for it. That is a hiring decision you get to not make, not a person you let go. If what you actually want is to cut the team you have, I am the wrong person to build it.
Is this just ChatGPT wired into my CRM?
No, and that distinction is most of the work. A language model sitting inside a linear automation runs the same play every time, regardless of what it produced last time. An agent reads the live data, decides on a cadence, and changes what it does next based on the result. The second one needs your data unified first, which is why it is a build rather than a plugin.
Do I actually need agents?
Probably not all of them, and possibly none of them yet. If the funnel upstream is leaking, an agent will automate the leak faster and make it harder to see. That gets checked in the build plan before anything is scoped. If the honest answer is that you need a hire, a pricing change or a fixed offer instead, I will say so.
What happens when an agent does something wrong?
Every agent gets a boundary before it gets autonomy: a spend ceiling, an approval step, a list of things it is not allowed to touch. Most start with a human approving the output and move to running on their own once the decisions have been right for a few weeks.
Anything that can spend money or publish publicly keeps a kill switch your team controls, and it is in the documentation, not in my head.
Will an agent get my ad account banned?
Not from being an agent. Accounts get restricted when the API is hammered for data it was never meant to serve, or when generated creative walks into a policy the account has already been warned about.
So the read path stays inside documented limits, and anything headed for publication passes a policy check before it goes live. On an account with a claims history, we keep a human approval step in place permanently. That is a deliberate trade of speed for not losing the account.
Can you add agents to a system somebody else built?
Usually. The real question is whether the data underneath is in good enough shape to make decisions from. Sometimes it is and the agent work starts immediately. Sometimes the tracking and the CRM disagree about what a lead is, and that has to be fixed first, or the agent optimizes toward a number that was never real.
Who owns the agents after the build?
You do, on the same terms as everything else. They run in your cloud, on your accounts and your API keys. You get the code, the documentation and a recorded walkthrough. Nothing runs on my infrastructure and there is no license you have to keep paying for. See the handover manifest for the rest of it.
Give Your Team The Hours Back.
Get your build planOr just email hello@handoff.systems and tell me which of the six is eating the most of your week.
Capped at three active engagements at a time.