Tom Morrison (@morrimoz) header
Tom Morrison (@morrimoz) profile picture

From Guam to everywhere πŸ‡¬πŸ‡Ί Building with AI, exploring business ideas, and traveling the world. Self-taught coder sharing the process. Businessβ€’AIβ€’Travel

San Diego, California
Joined July 2017
Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)Photo by Tom Morrison (@morrimoz)
I had an AI agent run my job hunt every morning for 2 months It scouted Upwork, LinkedIn, Dice and others, scored every job, drafted the applications, and nothing went out without my yes Today I landed an AI engineer contract So I open sourced the playbook. Drop it into any AI agent (grok bot, claude, codex, whatever you use) and it'll run your hunt too: github.com/morrimoz/job-scou…
23
this is the part to sit with. once you're plumbed into the client's data the build gets cheap. the real work is knowing which workflow to touch and who owns it. I run my business with a team of agents and that's where most of my time goes
I had an "oh shit" moment on Friday with Riggs. One of our current enterprise clients called us in the morning asking if we could automate a workflow for them. Typically, our Forward Deployed Engineer would go over and build the application for them and we'd charge $20-$50k annually for the application. But since we are already plumbed into their internal data systems, we were able to ask Riggs to build the automation and it just did it. An hour after they called us, the app was already built, and better yet, they can just do it themselves without our FDE having to be there. Now their employees will be able to build out their own automations and agents and extrapolate them throughout the organization on a shared operations system.
5
Small bug that drove me nuts this week. In webmoz's weekly review, if a caption was still saving, your first Approve click got eaten. Click again, feel dumb. Fixed. Edit it, approve once, it moves. The first click should just work
2
FABLE 5.2 IS F*CKING INSANE I gave it one prompt and a couple of notes. it made the whole launch film for my startup no screen recordings, no video model, no editor 3D in blender, motion in code, ElevenLabs for the voiceover I’ve been building webmoz solo for 2 years. this took about 6 hours (about 3 - 4 hrs to render) prompt if you want it
3
1
1
186
the prompt only told it what webmoz does. it did the story, the look and every shot itself webmoz is an AI co-founder. it works out your business, builds the site, makes and posts your marketing every week, then handles the jobs you give it webmoz.ai
1
11
reply "prompt" and I'll send you the exact one I used
4
same shape I like for client work one job, one approval, a step you can correct while it waits, and the result saved where you can check it if you've got a weekly process that eats hours, that's the kind of thing I build for people tommorri.com
13
I reckon most small business software ends up as jobs you hand off this week webmoz Command got a view that follows the work from the ask to the saved result, across whatever apps you've connected did the job get done, and can I see it. that's what I care about
13
asked webmoz for a week of posts in one sentence it proposed the platforms and post counts, I edited that and approved once the posts came back into growth studio review. I skip the weak ones, add media where needed, then confirm before anything gets scheduled
12
the worst thing an AI tool can do is lose the job halfway webmoz Command now keeps a job together from the ask to the result you see what it's doing, what it's waiting on and which app it's in. if it needs your OK, that stays in the job too
5
drawing 50 agents is easy. getting them to follow one rule is harder mine is draft only. nothing gets posted or sent without me thats the only reason I let a bot scout X for me every morning
the full social media marketing team from 0 to 50 AI agents in one visual: πŸ“‚ social media team ┃ ┣ πŸ“‚ niche research ┃ ┣ πŸ“‚ niche definition ┃ ┣ πŸ“‚ audience and offer ┃ ┣ πŸ“‚ leaders and competitors ┃ ┣ πŸ“‚ search phrase finder ┃ β”— πŸ“‚ double-meaning excludes ┃ ┣ πŸ“‚ competitor tracking ┃ ┣ πŸ“‚ competitor agent ┃ ┣ πŸ“‚ industry agent ┃ ┣ πŸ“‚ creator tracker ┃ ┣ πŸ“‚ top 50 company ranker ┃ β”— πŸ“‚ answer video planner ┃ ┣ πŸ“‚ outlier videos ┃ ┣ πŸ“‚ outlier finder ┃ ┣ πŸ“‚ niche match filter ┃ ┣ πŸ“‚ views + engagement ranker ┃ ┣ πŸ“‚ saves + shares ranker ┃ β”— πŸ“‚ dead views flag ┃ ┣ πŸ“‚ outlier creators ┃ ┣ πŸ“‚ creator finder ┃ ┣ πŸ“‚ engagement per 1K views ┃ ┣ πŸ“‚ two outlier minimum ┃ ┣ πŸ“‚ 30 day history flag ┃ β”— πŸ“‚ creator watchlist ┃ ┣ πŸ“‚ hooks ┃ ┣ πŸ“‚ hook bank ┃ ┣ πŸ“‚ structure grouper ┃ ┣ πŸ“‚ structure counter ┃ ┣ πŸ“‚ median views ranker ┃ β”— πŸ“‚ first film picker ┃ ┣ πŸ“‚ formats & trends ┃ ┣ πŸ“‚ format finder ┃ ┣ πŸ“‚ filming style mapper ┃ ┣ πŸ“‚ top vs bottom quarter ┃ ┣ πŸ“‚ copy this week ┃ β”— πŸ“‚ formats to copy ┃ ┣ πŸ“‚ scripts ┃ ┣ πŸ“‚ top 5 hooks ┃ ┣ πŸ“‚ script writer ┃ ┣ πŸ“‚ voice matcher ┃ ┣ πŸ“‚ source citer ┃ β”— πŸ“‚ next script loop ┃ ┣ πŸ“‚ scaling ┃ ┣ πŸ“‚ winner breakdown ┃ ┣ πŸ“‚ structure keeper ┃ ┣ πŸ“‚ concept remixer ┃ ┣ πŸ“‚ script per concept ┃ β”— πŸ“‚ most shared patterns ┃ ┣ πŸ“‚ reporting & briefs ┃ ┣ πŸ“‚ top 10 table ┃ ┣ πŸ“‚ Monday brief ┃ ┣ πŸ“‚ double down, stop, test ┃ ┣ πŸ“‚ five scripts ┃ β”— πŸ“‚ team report ┃ β”— πŸ“‚ content operations ┣ πŸ“‚ Claude connector ┣ πŸ“‚ Claude Code setup ┣ πŸ“‚ agent playbook reader ┣ πŸ“‚ agent reuse check β”— πŸ“‚ recurring agent (the one prompt that builds all 50 is in the article ↓)
8
agree, and the threads matter more than the model each of my bots has one lane and reports to a chief of staff bot, so I read one morning summary instead of ten chats same setup I'd bring into a client's slack
Slack is an especially good form factor for interacting with AI agents. Channels, threaded replies, notifications - all the features that make it work well for a team of humans work just as well for a team of agents.
14
spawn runs ten people and a fleet. I run solo with a handful of bots, chief of staff, research, marketing, a job scout, code manager and Claude & Codex the AI judges bit is what I'm stealing. turning my old calls into rules beats approving everything by hand
18
forward deployed agents still need someone forward deployed to run them I've got a team of bots at home. chief of staff, research, marketing, code, job scout etc. they draft and I approve mine only got useful once I wrote down what each one owns and when it should stop and ask
1
1
74
I'm bad at posting consistently. so webmoz writes my week, makes the images and videos, and I read it before it goes out. check out the demo:
18
Distribution is the best moat left. Anyone can build a feature in a weekend, what we cant copy is an audience that watched us build it Hoping this allows me to be consistent I suck at distribution. So I built a system instead:
πŸ€– Made with AI
Me thinking about my remaining moat
1
4
Then the same video becomes a short clip, a text post and an image post through the week. One build week becomes one anchor video and a week of posts. I just removed every step where I used to stall.
1
1
I turned this whole system into routines & skill files for Claude and Codex - kicked off and managed by Grok Bot lmk and I can share them with you if you want to run the same pipeline.
7