AI Process Visualization
Created by Ben Somekh

One person. A team of AI bots.
I approve everything that goes out.

Each bot has one job. They all read the same shared context, use the tools we already use, and hand work to each other. Anything external comes to me first.

Nothing external goes out without my approval
One job per botSpecialists, not one bot that does everything
Shared contextEvery bot starts from the same picture
A human approvesBots draft. I decide.
It learnsMy decisions and corrections make the next draft better
01
How I set this up

A team of bots, built in Grok Bot

Each bot is its own assistantI built this in Grok Bot, a desktop app. Each bot has its own job, chat, memory and schedules.
They share a foundationThe bots share one computer, one library of skills (reusable playbooks) and shared context files. They message each other and hand off work.
They connect to our toolsEach bot connects to Slack, Jira, Confluence, Figma, Gmail and more through connectors.
What's different from a single chat assistant like ClaudeThis is a team of always-on, specialized bots that run on schedules and talk to each other, not one chat you open when you need it.
What carries overClaude also supports reusable skills, projects with shared context, and connectors. So one job per assistant, shared context, written playbooks and human approval all work there too.
02
Meet the team

One job per bot

Think of it as a small product team. A chief of staff up front, specialists behind, and two bots that look after the team itself.

Chief

The router and chief of staff

My front door. I ask Chief, and Chief sends the work to the right bot.

  • Routes every ask to the right bot
  • Keeps shared context up to date
  • Keeps the glossary of people and terms
  • Runs release process reminders
  • Sends an end-of-day recap of what's pending
  • Keeps my to-do list tidy

Product Craft

PRDs, prototypes and product calls

Frames the problem, scopes the work, shapes the vision and pushes back when needed. Drafts and shows.

Execution Lead

Delivery

Keeps committed roadmap ideas and Jira tickets honest: status, owner, PRD and Figma links. Triages small fixes.

Bug Smasher

Tiny front-end bug and copy fixes

Writes the code and opens a pull request. An engineering lead reviews and merges. It never merges itself.

Product Research & Discovery

Competitors, customers, impact

Weekly competitor scan. Weekly customer signal scan from support tickets and sales calls. Partner engagement canvas. Measures features 30, 60 and 90 days after launch.

Product Enablement

Helping Support, Success and Sales

Support companions, go-to-market briefs, release FAQs, and answers to sales and support questions.

Meeting OS

Before and after meetings

Preps me before meetings. Turns meeting notes into to-dos afterward.

Shared with the team

Product Feature Request

The feature request board

A shared team bot. Keeps the board complete and current so people can decide.

Dad Bot

Parental-leave coverage

Keeps my coverage plan ready so the team owns every area while I'm out.

Alfred

Org designer

Designs and audits the bot team. Runs weekly health checks on how the bots are doing.

dr eggbot

Bot builder

Builds and fixes the bots. A new bot only when a real job needs one.

03
What goes in

Shared context, plus the tools we already use

Every bot starts from the same picture of our product, our people and how I like to work. Then each one connects to the tools its job needs.

Shared by every bot

  • Product contextOne shared file on our product and platform facts
  • Glossary of people and termsWho's who, and what our words mean
  • Memory of my rules and preferencesLearned once, applied everywhere
  • A library of reusable skillsPlaybooks like how to write a PRD, my writing voice, and thinking models
  • My to-do list in RoamOne list the bots read and keep tidy
Update it once, and every bot knows.

Connected tools

Talk and plan
SlackGmailGoogle CalendarGoogle Drive, Docs, Sheets, SlidesRoam Research
Build and ship
Jira and ConfluenceFigmaGitHub
Hear from customers
Granola meeting notesGong sales callsZendesk support
Data and research
Pendo analyticsDatabricksMongoDBLenny's Newsletter and podcastX
Bots work togetherThey hand work to each other, with a receipt for every handoff.
Bots run on schedulesLike a Monday competitor scan or a weekday morning meeting plan.
04
What comes out

Real work, in the places we already work

Outputs land in Jira, Confluence, Figma, GitHub, Slack and email. Anything external comes to me first.

Bot
Typical outputs
Before it goes out
Product Craft
PRDsPrototypesProduct decisions and pushback
I review every draft
Execution Lead
Jira ticketsTicket hygiene: status, owner, linksSmall fixes triage
I review first
Bug Smasher
Pull requests for small fixes
An engineer reviews and merges
Research
Weekly competitor digest in ConfluenceWeekly customer signal digest30, 60, 90 day feature checks
Auto-published internal digests
Enablement
Support companionsGo-to-market briefsRelease FAQs
I approve before publishing
Meeting OS
Meeting prepTo-dos from meeting notes
For me only
Chief
End-of-day recapRelease remindersA tidy to-do list
For me only
Every bot
Drafted Slack and email messages, waiting for my approval
I approve or rewrite
05
How it works

Bots draft. I decide. The system learns.

Every piece of work follows the same loop. And every loop makes the next one smarter.

Learning loop: my corrections become playbook fixes I approve 6 What shipped and what I decided is saved to shared context WHAT THE AI READS Shared context Product context People and terms glossary My rules and preferences Skills library of playbooks My to-do list in Roam Connected tools Slack and email Figma Jira and Confluence Data Meetings: Granola Customers: Gong, Zendesk, Pendo TRIGGERS I ask A schedule fires 1 Chief routes the work 2 Routes every ask. Keeps every bot up to date. SPECIALISTS DRAFT 3 Product Craft PRDs and prototypes Execution Lead Jira tickets, kept honest Bug Smasher Pull requests for small fixes Enablement Support companions and briefs And more Feature board, leave coverage Research Competitor and customer digests Meeting OS Meeting prep and to-dos APPROVAL I review and approve Nothing external goes out without my approval. Edit, push back or say no. 4 PUBLISHED OR SENT 5 Slack messages Emails Confluence pages Jira tickets Pull requests an engineer merges Under my name, with my sign-off SKIPS THE GATE Internal digests publish on their own Research posts weekly digests to internal Confluence For me only End-of-day recap, meeting prep, to-dos
1
I ask, or a schedule firesA Slack message to Chief, or the Monday competitor scan starts on its own.
2
Chief routes itChief reads the shared context and picks the right bot.
3
A specialist draftsUsing our tools and the right playbook.
4
I review and approveNothing external moves until I say yes. Internal digests and my own recaps skip the gate.
5
Published or sentTo Slack, email, Confluence or Jira. Code goes to an engineer to merge.
6
Saved for next timeWhat shipped and what I decided goes back into shared context. Chief keeps every bot up to date.
The learning loopA skill with its own schedule reviews what the bots did and where I corrected them, then proposes playbook fixes. I approve, and the system improves.
Drafts firstBots show me a draft before they post, publish or send.
Checks before publishingNo internal notes on shared pages. The bot checks the live page after it goes up.
A person merges codeBug Smasher opens a pull request. An engineering lead reviews and merges.
06
Try it yourself

Ideas you can borrow

You don't need a whole team of bots. Start with one job and these habits.

01
One job per botA bot with a clear job does better work than one that does everything.
02
Shared contextWrite down your product facts, people and terms once. Every bot uses them.
03
Reusable playbooksSave how you like a task done, like a PRD or a release FAQ, and reuse it.
04
Approve before anything externalLet AI draft. Read it, make it yours, then send it.
05
Scheduled routinesPut recurring work on a schedule, like a weekly scan or a morning plan.
06
Learn from correctionsGive it a skill and a schedule of its own. Weekly, or whatever fits, it reviews what the bots did and where you corrected them, then proposes playbook fixes. You approve them, and the system improves over time.
Want to try it? You can do much of this in Claude too.
Use skills for your playbooks, projects for shared context, and connectors for your tools.
AI Process Visualization ยท Created by Ben Somekh
Questions or want help setting up your own? Ask me.
07