A few weeks ago I posted a countdown. Fable 5 was about to go from free to paid, and I listed eight revenue systems you could still build for nothing before the meter started. The post did well, but the number that mattered to me was buried in the caption: lines of code written by me, zero. I run Growth Cab, a GTM advisory, and every one of those eight systems runs in production right now. None of them is a product I bought. Each one is a small AI sales workflow I described in plain English one evening and never had to touch again.
in
“Here are 8 revenue systems to build while it's still free. I run all 8 at my agency. Lines of code written by me: zero. Each one took a single evening in Claude Code.”
What an AI Sales Workflow Actually Is
When I say AI sales workflow I do not mean a chatbot you open and talk to. I mean a small program that does one job on its own. It wakes up on a trigger, pulls the data it needs, does the work, checks itself, and hands you a result. The reason these got cheap to build this year is that you no longer write the code. You describe the job to a model inside something like Claude Code, it writes and wires the thing, and you keep correcting it in words until it behaves. The skill that used to be engineering is now the ability to explain a process clearly. That is the whole change, and it is why a founder with no technical background can now run systems that used to need a team.
The Eight AI Sales Workflows I Actually Run
Here are the eight, grouped by the job they do. Three of them sit close to the customer. A call-prep brief reads a prospect's last ninety days of public activity and hands me a one-page summary before every demo, so I walk in already knowing what changed on their side. A follow-up writer turns a call transcript into a send-ready recap email in my voice. An outbound researcher scans my account list every week and finds the one fresh trigger worth reaching out about for each name, so nobody gets a message that opens with nothing to say.
Three more give me visibility. A CRM dashboard refreshes my pipeline numbers every night while I sleep, so the first thing I see each morning is the real state of the quarter. A KPI board shows every rep the single number they should move that day. A ROI calculator sits ready to screen-share live on a call, so when a prospect asks whether the math works I show them the answer instead of promising it.
The last two are pure back office. A payment watchdog reconciles our invoices against the bank every Friday and flags anything that did not land. An inbox triager reads the overnight pile and drafts replies in my tone, so I approve instead of type. Each of these took one evening to build. Not one of them carries a monthly seat fee, because the tool belongs to me.
Why Building an AI Sales Workflow Beats Buying Another Tool
The obvious objection is that you can buy most of this. There is a vendor for pipeline dashboards, a vendor for follow-up emails, a vendor for account research, and I have paid for plenty of them. The problem is never that they do not work. It is that each one solves eighty percent of my exact problem and charges me a seat for the privilege, and stitching six of them together becomes its own full-time job. When I describe the workflow myself, it does exactly my process, touches my data, and speaks in my language, because I built it around how Growth Cab actually sells rather than how a product manager guessed the average company sells. The running cost is API credits measured in single-digit dollars a month. And when my process changes, I change the workflow in a sentence instead of filing a feature request and waiting a quarter.
Where AI Sales Workflows Still Break
I am not going to pretend these run themselves forever. The place they break first is verification. A workflow that researches an account will, given a thin profile, fill the gap with something fluent and wrong. I once had one write a warm opener around a funding round that never happened, and a prospect replied to correct us. The fix was a rule: every factual claim has to trace back to a source the workflow actually pulled in this run, or it does not get said. That single check killed almost all of the errors, but I had to know to write it. A workflow with no verification step is not autonomous. It is unsupervised, and those are very different words.
The second place they break is data. A follow-up email or a research brief is only ever as good as what you feed it, so a sparse CRM produces confident filler. The third is judgment. None of these systems can read that a deal went quiet because the champion is on leave rather than because they lost interest, so they keep nudging at exactly the wrong moment. And there is a scale problem worth naming out loud. A human sending one wrong email is a bad morning. A workflow sending four hundred wrong emails is a brand problem. That asymmetry is why anything touching a live deal still has a person on the approval step at Growth Cab. The workflow proposes and I approve, and only the low-risk slice runs fully on its own.
How to Build Your First AI Sales Workflow This Week
Do not try to automate your funnel. Pick one narrow job you already do by hand every week and quietly hate. For most founders that is call prep or follow-up. Open Claude Code, describe the job the way you would explain it to a new hire, and let it build the first version. Then run it in draft mode for a week, where it proposes and you approve every output by hand. You will find the holes fast, and each hole becomes a rule you add in one sentence. Once it stops surprising you, let it run on the safe slice and keep yourself on anything with revenue attached. The honest math is that my eight workflows save me a real chunk of every week, but only because a person still owns the send on the things that matter. The teams winning with AI sales workflows are the ones who moved the human from doing the work to approving the loop.
If you want the built version of these, I break down one GTM workflow a day in The Revenue AI Brief, my newsletter, with the trigger and the verification rule written out in full. The original post that kicked this off, with all eight systems listed, is on my LinkedIn. Come tell me which one you would build first, or where you think I am wrong about keeping a human on the send.
