Is this too basic for me?

Probably not. You join data from two sources, write a system spec, build triggers, and ship an agent with an approval step and prompt-injection defenses. Then you add errors, alerts, logs and limits. We pair people by level so everyone builds at their own pace.

Do I need to know how to code?

No. Registration asks which AI tools you use and your level. No minimum level is set. Your answers set the pace and your build partner.

Will it work with my software?

If you can get your data out with a Composio connection, a CSV export or a login, yes. You connect it during pre-work, before session 1.

My software has no connection for other tools. Can I still do this?

Yes. Session 2 covers Browserbase and Doppler for systems with no API, including what to check in your vendor terms. A CSV export is the fallback.

When are the sessions?

Wednesdays 28 October, 4, 11 and 18 November, 90 minutes each. We set the time on 14 October from the times registrants pick, and email everyone the same day.

What happens before session 1?

Pre-work from 22 to 27 October: accounts, a read-only data connection, keys in Doppler, and a Claude Project with your context doc. Three short videos walk you through it. Office hour on 26 October if you get stuck.

Can two people from my business share a seat?

No. A seat is one person. Up to 3 people from one company can join. A team of 3 is $2,400.

What if the cohort does not fill?

We decide on 21 October. With fewer than 4 paid seats, the start date moves and your seat moves with it.

Cohort #1Live · online · 4 Wednesdays

Build the tools your software can't give you.

In 4 weeks, on your own data.

A live, hands-on cohort for owners and their teams. You join your data, build a report and an agent that acts safely, and hand the system to your staff. You leave with one working system and a written spec for the next one.

DATES: 28 Oct to 18 Nov
SESSIONS: 4 × 90 min, live
GROUP: Up to 10 people
PRICE: $800 per person

PRE-WORK 22–27 OCT · CLOSES 21 OCT · RUNS WITH 4+PRE-WORK 22–27 OCT · REGISTRATION CLOSES 21 OCT · RUNS WITH 4 OR MORE

The problem

Your software does most of the job. The rest gets done by hand.

Your business runs on one main piece of software. It does most of the job. The rest gets done by hand, every week, by you or your staff. This cohort teaches you to build the rest yourself.

MOST OF THE JOBYour software

THE RESTBy hand, every week

What you leave with

One working system on your own data. And a plan for the next one.

TOOL 01 · CAPTURE

A form that captures what your software does not.

Each entry fires a trigger, so the next step starts right away.

TOOL 02 · REPORT

A scheduled report that combines your system data with what you capture.

A PDF in your inbox every week. It also measures what the agent did.

TOOL 03 · AGENT

An agent that drafts follow-ups, reminders and reorder lists.

You approve before anything sends. It gets only the tools it needs and treats every input as data.

GRADUATION · PART 1

System 1 shipped.

The three tools chained into one system, packaged as Claude Projects or Skills, tested, handed off to your staff and demoed to the group.

GRADUATION · PART 2

A spec for system 2, written by you.

Due 25 November. You get written feedback on it. Optional check-in on 2 December to show what you built.

Week by week

One system, built in layers.

Wednesdays, 90 minutes, live and online. We set the time on 14 October from the times registrants pick.

00
PRE-WORK · 22–27 OCT
Set up before day 1
- Accounts, data connected read-only, keys in Doppler
- A Claude Project with your context doc
- 3 short videos, office hour on 26 Oct
01
WED · 28 OCT
Map, model, spec
- Map one process
- Join two data sources on a shared ID
- Write the spec for system 1
02
WED · 4 NOV
Capture, triggers, report
- Tool 1: capture form with a trigger
- Tool 2: scheduled PDF report
- Systems with no API: Browserbase and Doppler
03
WED · 11 NOV
An agent that acts, safely
- Tool 3: agent with an approval step
- Defend against prompt injection
- Package it as a Skill your staff can run
04
WED · 18 NOV
Harden, hand off, demo
- Errors, alerts, logs and limits
- Access review and staff hand-off sheet
- Demo, then start the spec for system 2

Most of each session is building

10 min wins, 20 min concept and live demo, 45 min building on your own data, 15 min blockers and homework.

Setup happens before day 1

Pre-work from 22 to 27 October. Done when Claude can read your data. Session 1 starts with building.

Help between sessions

Every session recorded and sent within 24 hours. One open office hour each week, and a group chat.

Paired by level

Up to 10 people. You tell us your AI tools and level when you register. We set the pace and the build pairs from your answers.

Fit

Is this for you?

It is for you if

You run on one main software package and your team works around it by hand.

You, or someone on your team, can give 3 hours a week for 4 weeks, plus the pre-work.

You can get your data out: a connection, an export, or a login.

You want to build on your own data. There is no sample track.

Any level. From people who use AI chat now and then to people who write code and build agents. No coding needed.

It is not for you if

You want it built for you. That is a First Build.

You cannot get to your own data.

You want to learn servers, databases or data sync. We show what they do. We do not teach them.

Want it built for you instead?

Talk to us for 30 minutes

What you learn

The skills behind a system that keeps working.

Everything runs in your own accounts. You keep what you build.

Claude Projects and Skills
Composio
Doppler
Google Sheets and Gmail
Browserbase

Joining dataTwo sources matched on a shared ID, such as a customer or job number.

System specsWhat starts it, inputs, outputs, approvals, failure paths, how you measure success.

Triggers vs schedulesWhen a form entry should act now and when a report should run on a clock.

Safe agentsApproval steps, least tools, and defenses against prompt injection.

Testing5 test inputs per tool, including one hostile input.

Running itErrors and alerts, run logs, usage caps, a clear stop, and an access review.

WHAT WE DON'T TEACH

Servers, databases, or keeping a daily copy of your data.

We show you what they do. If you need them, that is what STOA builds.

AFTER THE COHORT

Want to keep building it yourself?
STOA can coach you.

AI build coaching for you or one person on your team, or in-house training for your whole team. You do the building. We guide each step and check the work.

Ask about coaching and training

What it costs

One price per person. No upsell in the room.

Up to 3 people from one company. Each person is one seat.

REFUND POLICY

Ask for a refund by 18 October, 10 days before session 1, and you get it in full. No questions asked. After 18 October, there are no refunds.

$800PER PERSON

Team of 3 from one company$2,400

TOOLS YOU PAY FOR YOURSELF

Claude Pro: $20 a month
Composio, Doppler: Free tiers
Google account (Sheets, Gmail): Free
Browserbase, only if you have no API or export: Free tier

Who teaches it

Alejandro Morales

Alejandro is the founder and COO of STOA, the software team for small businesses that don't have one. STOA decides what gets built on a business's software and then builds it. In this cohort, he teaches you to build part of it yourself.

About STOA What clients say

Questions

Before you register

Is this too basic for me?: Probably not. You join data from two sources, write a system spec, build triggers, and ship an agent with an approval step and prompt-injection defenses. Then you add errors, alerts, logs and limits. We pair people by level so everyone builds at their own pace.
Do I need to know how to code?: No. Registration asks which AI tools you use and your level. No minimum level is set. Your answers set the pace and your build partner.
Will it work with my software?: If you can get your data out with a Composio connection, a CSV export or a login, yes. You connect it during pre-work, before session 1.
My software has no connection for other tools. Can I still do this?: Yes. Session 2 covers Browserbase and Doppler for systems with no API, including what to check in your vendor terms. A CSV export is the fallback.
When are the sessions?: Wednesdays 28 October, 4, 11 and 18 November, 90 minutes each. We set the time on 14 October from the times registrants pick, and email everyone the same day.
What happens before session 1?: Pre-work from 22 to 27 October: accounts, a read-only data connection, keys in Doppler, and a Claude Project with your context doc. Three short videos walk you through it. Office hour on 26 October if you get stuck.
Can two people from my business share a seat?: No. A seat is one person. Up to 3 people from one company can join. A team of 3 is $2,400.
What if the cohort does not fill?: We decide on 21 October. With fewer than 4 paid seats, the start date moves and your seat moves with it.
What is the refund policy?: Ask for a refund by 18 October, 10 days before session 1, and you get it in full. No questions asked. After 18 October, there are no refunds.

Save your seat in Cohort #1.

Your answers set the session time, the pace and the build pairs. We use your process to shape the examples.

4 WEDNESDAYS · 28 OCT TO 18 NOV
PRE-WORK · 22 TO 27 OCT
UP TO 10 PEOPLE · $800 EACH
REGISTRATION CLOSES 21 OCT