Birostris ✦

For solo founders: I show how to set up AI agents for planning, inbox, content, research and admin with copy-paste prompts. No code. You approve every draft.

[Birostris ✦ @Birostris_77]

One time purchase, own the system forever:
https://t.co/O1uSEosxze

[Birostris ✦ @Birostris_77]

Pyramid Transport & Cold Storage ships truckloads of lobster and snow crab worth up to $800,000. One of its best fixes had nothing to do with the trucks. It was the end of every meeting.

CTO Steve Swain told Google: "Before, we'd leave a meeting confused about who's responsible for what."

Now Gemini in Google Meet sends out the notes in seconds once the meeting ends.

What the write-up describes:
• A seafood shipping company started in 1993 by a former truck driver
• Notes and step-by-step procedures go out right after meetings
• Leaders keep backup plans in one place staff can ask questions of
• Steve's word for the change: accountability

The notes matter less than one line inside them: a name next to every task.

My version, with an admin agent:
1. I paste in the notes or the transcript.
2. It lists every task with one owner and a due date.
3. Anything without an owner gets marked [WHO?].
4. I check it, then I send it.

My own setup is copy-paste prompts for five agents: planning, inbox, content, research and admin. Written to work in Claude, ChatGPT or Grok. Details in the reply below ↓

How do your meetings end right now? Comment below.

[Birostris ✦ @Birostris_77]

@gregisenberg Same with AI agents. I keep using the ones where I wrote the rules myself, not the copied setups. Do you think AI products should make users build part of the setup on purpose?

[Birostris ✦ @Birostris_77]

@WSterlingSpace Fair point. I'd have the agent mark every number or "maybe" it heard with [CHECK], so the scan only takes a minute. Do you price from that list yourself, or let the tool fill in rates?

[Birostris ✦ @Birostris_77]

Rob Castelao runs Good Guy Plumbing in Torrance, California: 12 full-time people and eight vans. He told CONTRACTOR Magazine his three office staff free up roughly 10-plus hours a week with AI tools, about 500 hours a year.

The part I'd copy is the approval step.

What the Sept 2025 write-up describes:
• An AI voice agent answers after-hours calls, triages the issue and books the job on the calendar
• Scheduling can suggest a tech, a time and a route
• Coach AI drafts a line-item quote from their past labor and materials
• A tech approves that quote on site in about 90 seconds
• The dispatcher or field lead still has final say. AI suggestions stay suggestions

He also spent a couple mornings role-playing calls with the team before trusting it. Fast onboarding. Slow permission.

How I'd copy that on my own:
1. Pick one job that keeps stealing evenings, like after-hours intake or quote drafts.
2. Let AI write the first version only.
3. Keep a human override on anything a customer will see or pay for.
4. Review the early runs yourself for a full week before you loosen the leash.

My own setup is copy-paste prompts for five agents: planning, inbox, content, research and admin. Written to work in Claude, ChatGPT or Grok. Details in the reply below ↓

[Birostris ✦ @Birostris_77]

https://t.co/hSKNEiyW64 writes about four case studies a month. With Gemini's help, the team turns them into around 80 videos and 160 social posts.

CEO Nic Breedlove sells play structures to schools, parks and cities, around 6,000 customers a year.

The math from the Google customer story:
• 4 case studies a month
• About 80 videos made from them
• About 160 social posts
• So roughly 20 videos and 40 posts per story

The part I like is the starting point. Every post comes from a real job they already finished, not a blank page.

Next time you finish a client job, do this:
1. Write the story once: the problem, what you did, what changed.
2. Paste it into your content agent with your real notes and photos.
3. Ask for 5 short posts, each from a different angle.
4. Ask for 2 video scripts under 60 seconds.
5. Cut any line that claims something the story doesn't say.
6. You pick what goes out. It never posts on its own.

The DM gives you The AI Team System: AI agents for planning, inbox, content, research and admin, set up with copy-paste prompts, no code, works with Claude, ChatGPT or Grok.

Which finished job of yours deserves 10 posts instead of zero?
Comment AGENT and I'll DM you the details.

[Birostris ✦ @Birostris_77]

@dylan522p Interesting. The agents that stuck for me are the ones with one standing job, like sorting my inbox every morning, instead of the ones I open a chat for. What made Perplexity Computer stick for your team, the harness or the mix of models?

[Birostris ✦ @Birostris_77]

@emollick True, that aged fast. For my small business I skipped the search setup and keep one short written file of my prices, rules and tone that every agent reads first. Where do you think retrieval still earns its keep, very large document piles?

[Birostris ✦ @Birostris_77]

The Boardwalk Cleaning Co. has 42 employees who clean 13,000 homes a year in Austin. When a cleaner needs to know how to safely clean a marble surface, they can ask the company's own 50 pages of how-to steps.

Founders Jessica Duarte and Paula Jones put their standard operating procedures into Gemini Notebook. The team asks in plain words and gets the answer from those pages.

What else the Google customer story mentions:
• The sales team keeps its playbooks in there too
• Jessica has Gemini build the pivot tables she used to make by hand
• The company has been running since 2003

Most small businesses already have this knowledge. It just lives in one person's head.

You can start yours this week:
1. Write down the 10 questions new people ask you most.
2. Answer each one in plain words, one page or less.
3. Give your research agent that file and one rule: answer only from it, and say "not in the doc" when it can't.
4. Every "not in the doc" is your next page to write.

What question do people ask you so often you could answer it in your sleep? Comment below.

[Birostris ✦ @Birostris_77]

@businessbarista Useful split between projected and realized value. For a one-person business I'd track a smaller version: how many AI drafts go out each week with only light edits, since heavy rewrites eat the saved time back. Would 1-3-1 work at that size, or is it built for bigger teams?

[Birostris ✦ @Birostris_77]

@Yuchenj_UW Same here. In a small business I write the context down myself: prices, rules and tone in one short file every agent reads first. Does Genie still need someone to correct what it learns about how the business works?

[Birostris ✦ @Birostris_77]

@omarsar0 Agreed. I'd test a dictation tool on one real customer email and check that names, dates and prices survive my mid-sentence corrections. Does the bench count a changed number as worse than a changed word?

[Birostris ✦ @Birostris_77]

FR8relay uses Gemini to read 3,000 broker emails a day. It turns messy rate quotes into short bullet points and drafts the replies.

Aayush Thakur and his wife, Deme Yuan, built the company so truck drivers can get home daily. Their software splits long routes into relays, so drivers swap trailers and head back.

Per the Google customer story:
• About 20,000 loads a year run through their platform
• Long rate quotes become clear bullet points
• Replies get drafted to keep deals moving

The word I'd underline is "drafted." A wrong rate in a reply costs real money, so I'd still want a person reading it before it goes.

Copy these rules if your inbox is full of quotes and back-and-forth:
1. Put a 3-line summary on top of every long thread: who, what they want, the deadline.
2. Pull every price, date and quantity into one list.
3. Draft the reply. Never send it.
4. Flag any number that doesn't match the last email in the thread.
5. Read it, fix it, then you hit send.

The DM gives you The AI Team System: AI agents for planning, inbox, content, research and admin, set up with copy-paste prompts, no code, works with Claude, ChatGPT or Grok.

What's the email you answer most often each week?
Comment AGENT and I'll DM you the details.

[Birostris ✦ @Birostris_77]

Rich James runs Lucerne Farms in far northern Maine. He gave Gemini his list of New England stores and asked for the best sales routes. It came back with three weeks of trips to around 400 stores, seven to eight a day.

His team moved the routes into Google Maps. Now his sales reps start each day with the stops already planned.

A few details from the Google customer story:
• The farm harvests 6,500 to 7,500 acres of hay and straw a year
• 15 people work there year-round, 27 at harvest
• Its products are in 2,000+ garden centers and stores
• The AI also suggested the best time of day to visit each store

To me, this is the best kind of first job for a planning agent. The output is easy to check on a map before anyone drives anywhere.

Try it on your own week:
1. Paste in every place you need to be, with addresses.
2. Add the fixed times: school pickup, calls, opening hours.
3. Ask for a day-by-day plan with travel time between stops.
4. Ask it to list anything it wasn't sure about at the bottom.
5. Check the plan on a map before you trust it.

Which weekly schedule or route would you hand to AI first? Comment below.

[Birostris ✦ @Birostris_77]

@WSterlingSpace Makes sense. I'd have the agent turn the transcript into a plain materials list first, then write the quote from that list, so the guessed lines are easier to spot. Do voice notes ever miss things your typed ones would catch?

[Birostris ✦ @Birostris_77]

@LMJCFO That's a useful order. I'd put it in the agent's instructions as step one: check unapplied payments and credit memos against the aging, then write the summary. Do you tie it out by hand each month, or does your system flag it?

[Birostris ✦ @Birostris_77]

Dan Rosenthal runs a 22-person services company. He turned its how-to docs into more than 50 sets of instructions for Claude.

The part I'd copy is the rule he keeps next to them: 100% AI output isn't acceptable. Everyone owns their work.

What the setup does, per his post:
• Each process doc became a plain file the AI reads first
• Most of them now do about 70% of the task, he says
• Any of his 22 people can start an assistant with all of it in seconds
• Every project has a safeguard file that blocks certain actions
• When someone corrects the AI, the correction gets fed back in

Those are his numbers, not mine. His team also runs 30+ connected tools, which a solo owner doesn't need.

The idea works at any size, though. The AI is only as good as what you've written down for it.

The small version, for one person:
1. Pick the job you explain most often, like answering a new customer.
2. Write the steps the way you'd tell a new hire.
3. Add a short list of things it must never do.
4. Paste both in before you ask for a draft.
5. When you fix a draft, add the fix to the steps.

The DM gives you The AI Team System: AI agents for planning, inbox, content, research and admin, set up with copy-paste prompts. No code. Written to work in Claude, ChatGPT or Grok.

Which job would you write down first?
Comment AGENT and I'll DM you the details.

[Birostris ✦ @Birostris_77]

One time purchase, own the system forever:
https://t.co/O1uSEosxze

[Birostris ✦ @Birostris_77]

A gardener in England records a voice memo between jobs, and AI turns it into an invoice. Each request costs less than a cent, says the man who built it.

Nadia Dewhurst-Richman runs Wee Beastie Gardens. Her husband, Danny Richman, a consultant, built the setup. Zapier wrote it up in May 2024.

How it works, per that write-up:
• She records a quick memo on her phone
• AI works out if it's an invoice, an expense or a task
• It pulls out the client, the service and the amount
• In their version, the invoice goes out on its own

The write-up itself says you can add an approval step. I'd add it on day one. A wrong amount on an invoice costs more than the minute it takes to check.

My version of the same idea:
1. Record the job notes while they're fresh.
2. The admin agent drafts the invoice from them.
3. I check the client, the amount and the date.
4. Only then does it go out.

My own setup is copy-paste prompts for five agents: planning, inbox, content, research and admin. Written to work in Claude, ChatGPT or Grok. Details in the reply below ↓

Which job would you record a memo for first? Comment below.

[Birostris ✦ @Birostris_77]

@ideabrowser Good stack. Owning the audience is the one I'm still building, so right now I answer every reply before I post more. Do you think owning outcomes can grow the audience, or does the audience have to come first?

[Birostris ✦ @Birostris_77]

In a Nov 4, 2026 customer story, Sahar Naim's inbox went from 30,254 emails to 0. Her agents did the cleanup. She still makes the decisions.

She runs ops for startups while finishing school, coaching her son's soccer team and co-parenting two teens.

It's a vendor write-up, so the numbers are theirs. Here's what she hands off:
• One agent cleans up her inbox every morning and keeps her week on track
• One is a sounding board for the business side
• One helps with her business law class
• One handles the kids' schedules and family appointments

Her own words: "I still make every call."

That line is the whole setup. The agents take the legwork. The calls stay with her.

Steal these before you hand an agent your inbox:
1. Start with the morning cleanup, not replies.
2. Have it sort mail into three piles: needs me today, can wait, archive.
3. It drafts replies. It never sends one.
4. Anything about money, contracts or family gets flagged for you.
5. Check its sorting for one week before you trust it.

The DM gives you The AI Team System: AI agents for planning, inbox, content, research and admin, set up with copy-paste prompts, no code, works with Claude, ChatGPT or Grok.

Which pile in your inbox would you hand off first?
Comment AGENT and I'll DM you the details.