Each film, then its full copy set: LinkedIn variants (one Recommended), the first comment, and the X main tweet + thread. Draft for your approval. Nothing scheduled or posted.
1 · Granola → HubSpot flagship
27.3s
After a call, the notes sit in Granola and the CRM stays empty. One Floom worker turns every call into an updated HubSpot deal, one review before it saves.
LinkedIn4 variantsplain voice, no hashtags
Variant A - his own workerRecommended
His worker, the 7pm pain, one honest admission.
I run a worker that updates HubSpot after every sales call.
It reads the Granola notes and fills in the deal fields: stage, next step, budget, champion. It gets the next step wrong sometimes, so I read it before it saves and fix that line when it does.
Before this I did it by hand around 7pm, or I'd be halfway through dinner and realize I never logged the last call. Stages sat stale for days and I'd forget who the champion even was.
I set the worker up on Floom and told it the job in two sentences.
Variant B - the admin I skippedAlternative
The task everyone half-skips, off his plate.
Updating HubSpot after calls was the admin I skipped most.
Now a worker does the first pass. It reads the Granola notes and fills the deal fields: stage, next step, budget, champion. It waits for me to approve before anything saves, and I usually tweak a field or two.
The whole thing is on Floom, no code and nothing to keep wired together.
I read the deal, fix what's off, and approve it.
Variant C - the last hourAlternative
Opens on the hour lost at the end of the day.
For a while the last hour of my day was copying calls into HubSpot.
Granola had the notes. Getting them into the right fields was the boring part, and at the end of a long day I let it slide.
A worker on Floom does it now. After each call it reads the notes and sets stage, next step, budget, and champion. I read it before it saves.
I don't do it at 7pm anymore.
Variant D - cold openAlternative
A specific lived moment, straight in.
More than once I've finished a call, meant to log the next step, and just didn't.
The worker I built on Floom reads the Granola notes right after each call and sets stage, next step, budget, and champion. It's wrong often enough that I read every one before it saves. That's the part I still do.
Before this, those details lived in my memory until 7pm, and plenty got lost.
First commentlink on comment, not post, keeps the reach
Link: floom.dev
You connect Granola and HubSpot and tell the worker what to fill in. After each call it updates the deal fields and holds for your approval before saving.
I built it for my own sales calls, so if you want to set up the same thing, ask me.
Xmain tweet + thread
Main tweet (no link in the main tweet)
i mostly stopped updating hubspot after calls. doing it by hand at 7pm never really happened.
now a worker reads the granola notes and fills the deal fields after each call. i read it and fix the odd line before it saves.
Thread
1/ it sets stage, next step, budget, champion. it's wrong sometimes, so nothing saves until i look.
2/ it's all on floom. no code, nothing to keep wired up.
3/ i approve it once and move on. floom.dev
2 · gmail-digest how it works
29.2s
A real worker, not a scheduled prompt. It reads Gmail each morning, writes a short digest, posts to Slack, and never sends in his name. The clip shows one being built.
LinkedIn4 variantsplain voice, no hashtags
Variant A - his real inbox workerRecommended
Closest to how he actually talks about it.
A worker reads my inbox every morning before I do.
It goes through Gmail, writes a few lines on the emails that want a reply from me, and posts that to Slack. It doesn't reply to anyone on its own.
It still flags things I don't care about and misses the odd one, so I open Gmail later anyway. But I'm not starting the day scrolling past cc's and newsletters looking for the three that matter.
I set it up on Floom by telling it what I wanted checked.
Variant B - the buildAlternative
Matches the clip: one being built, step by step.
The clip is me setting up an inbox worker on Floom.
I wanted my mail triaged before I open it, so I gave it these steps:
- read Gmail each morning
- flag the emails that want a reply from me
- write a few lines on each
- post that to Slack
- don't reply to anything itself
I didn't write any code. I use the same setup for a daily lead summary too.
Variant C - why I trust itAlternative
The approval is why he lets it near his email.
I was careful about letting anything automated touch my inbox.
The worker I run on Floom reads Gmail each morning, writes a short summary of what wants a reply, and posts it to Slack. It can't send or reply as me. It's set up that way on purpose, which is why I'm fine letting it read my mail.
The summary isn't everything, so I still open Gmail after I read it.
Variant D - the sorting hourAlternative
Opens on the morning triage.
My morning used to start with sorting email, deciding which ones mattered and which could wait.
A worker does that pass now. It reads Gmail, writes a short summary of what wants a reply, and posts it to Slack before I'm at my desk. It gets the priority wrong here and there, so I treat it as a first pass and still check myself.
I read the summary, then open the few that actually need me.
First commentlink on comment, keeps the reach
Link: floom.dev
You connect Gmail and Slack and tell the worker what to look for. It posts a morning summary and never sends mail on its own.
If you work in the terminal, you can add it to Claude Code:
npx @floomhq/floom mcp install
Xmain tweet + thread
Main tweet (no link)
a worker reads my inbox every morning before i do.
it summarizes the emails that want a reply and drops that in slack. it can't send mail as me, so the replies are still mine.
Thread
1/ i set it up on floom by telling it what to check: read gmail, flag what wants a reply, summarize it, post to slack, don't reply to anyone.
2/ it misses things and flags stuff i don't care about, so i still open gmail later. but i start from the summary instead of the full inbox.
3/ the clip is me setting it up. floom.dev
3 · Your agent, covered the abstraction
27.7s
Writing an agent is easy. Keeping one running is the mess: schedule, retries, credentials, approval, a machine to run on. That is the part Floom does.
LinkedIn4 variantsplain voice, no hashtags
Variant A - the folder of dead agentsRecommended
Specific agents he built, specific ways they broke.
I have a folder of agents I built and don't use. One pinged me when a competitor tweeted. One drafted replies to support mail.
They worked when I ran them by hand. Then I'd close my laptop and they'd stop. A tool would time out and nothing retried. My API keys were sitting in a script in plain text. One tried to email a contact twice, with no approval step to catch it before it sent.
I never got around to building any of that properly.
My inbox and CRM workers run on Floom now. I wrote what they should do, and Floom handles the schedule, the retries, the keys, and the approval step.
Variant B - the five placesAlternative
Names the exact gap Floom fills.
Every agent I stopped using broke in one of five places: the schedule, a tool with no retry, credentials I'd left in a script, no approval step, or nowhere to run once my laptop was shut.
That's the boring part around an agent, and I never kept it up. So I moved it onto Floom and wrote what the workers should do.
My inbox one and my CRM one run there now.
Variant C - broke in a weekAlternative
The build was fast; keeping it alive was the cost.
The first time I built my inbox worker, it broke within a week.
It needed a schedule, retries when a tool failed, somewhere safe for the credentials, an approval step, and a machine that stays on. I'd wired that up before for other agents and didn't want to do it again for this one.
So I rebuilt it on Floom, which handles all of that. I describe what the worker should do and it runs, with a summary in Slack every morning.
Variant D - the list, numberedAlternative
The plumbing list, and which parts he skipped.
This is the setup my agents needed to run on their own, and the part I kept skipping:
- a schedule that fires on time
- retries when a tool fails or an API drops
- credentials kept out of the code
- an approval step so it doesn't act in your name by mistake
- a machine that runs it when your laptop is off
I used to build that list myself for every agent and skip half of it, which is why most of mine stopped working.
Floom handles that list for me now. I write what the worker should do, and it runs.
First commentlink on comment, keeps the reach
Link: floom.dev
You write what the worker should do and Floom runs it: the schedule, the retries, the credentials, the approval step, and the machine it runs on. You set it up in the browser, no code needed.
If you work in the terminal, add it to Claude Code:
npx @floomhq/floom mcp install
Xmain tweet + thread
Main tweet (no link)
i have a folder of agents i built and don't use. one pinged me when a competitor tweeted, one drafted replies to support mail.
they'd stop when i closed my laptop, or a tool would time out with no retry, or my keys were sitting in a script. i never built the setup that keeps one running on its own.
Thread
1/ my inbox and crm workers run on floom now. i wrote what they should do, floom handles the schedule, retries, keys, and the approval step.
2/ i haven't had to rebuild any of that setup for them, which is what always killed the ones i made before. floom.dev