Brian MacDonald (@BmacBuilds) header
Brian MacDonald (@BmacBuilds) profile picture

Entrepreneur | Surfer | Founder @Ownboxio Everybody’s building AI agents, I built the place where they work.

Orange County, CA
Joined March 2021
Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)Photo by Brian MacDonald (@BmacBuilds)
How do you actually build a company brain? Here's mine. My business runs on one AI Brain: over a hundred documents with our offers, prices, policies and decisions. • One Start here page that every person and agent reads first • Folders by job, plus a Public folder • One topic per document: each heading a question, the first sentence the answer • A decisions log: newest first, one line each, with who decided • Agents connect through MCP. They read, and they propose changes I approve from my phone • The inbox comes in too. AI drafts the reply, and a person sends it No copying what one tool knows into the next. Nobody explaining the business twice. Part two of my series, the full build: brian-macdonald.com/guides/h…
1
62
A ChatGPT business agent only works inside ChatGPT. Your customers don't. Some ask ChatGPT. Some ask Gemini because it's on their phone. Some ask Claude or Perplexity. And every AI company is building its own. Google has a Business Agent too, inside Search's AI Mode and the Gemini app. An agent you build in one doesn't show up in the other. So you have two choices: • Build an agent in every AI app, and keep them all in sync by hand • Keep one original that every AI can read, and give each app a copy I keep one original, on a server my company owns. Your customers don't live in one AI. Your answers shouldn't either. Why one agent isn't enough: brian-macdonald.com/articles…
1
2
85
ChatGPT now has a listing for your business. Claim it this weekend. It's called Business Hub. It opened in beta on November 10, and claiming is free. What you get: • You check and correct what ChatGPT shows about your business • You can add a product feed with current prices • You see totals: mentions, clicks and common topics What you don't get: the conversations. And it only covers ChatGPT. So claim it, and fill it in from your own written answers, not from memory. Then keep the original in a place you own. Claim the listing. Keep the original. What Business Hub is, and what it isn't: brian-macdonald.com/articles…
3
1
75
The best-run businesses don't have more AI tools. They have one brain every AI works from. This is Ownbox: a shared company brain with a business agent built in, on a server you own. Get your free AI visibility report at ownbox.io/ai-visibility
1
84
Your AI doesn't know your business until you give it the answers. Here are the five documents I'd start with. Each one works in any tool: 1. The last five questions. Find the person everyone asks, and write down the last five questions they answered. 2. A decisions log. What you decided, who made the call, when and why. 3. The facts your AI gets wrong. Ask it what you charge and what your refund policy is. Write down the right answers, then ask again. 4. One brief per client. The job, the offer, what's approved and what changed. 5. A public page of answers. Who you help, what you sell and what it costs, written so AI assistants can find it. Five documents, and your AI stops guessing. The full guide: brian-macdonald.com/guides/e…
1
57
What is a company brain? It's one place that holds what your business knows: your offer, prices, policies, playbooks and best answers. Your team and every AI agent read from it and add to it. Here's why every business needs one now. Every AI tool you add knows one piece of the business. The chat assistant knows what you pasted this morning. The inbox tool knows your emails. The CRM knows your deals. None of them know what the others know. So your people become the glue. They re-explain the business to every new chat and catch the mistakes before customers do. A company brain fixes that: • Agents read it, not just people • Agents can add to it, with a person's yes • It stays one place, not five copies drifting out of date I run my own company on one. This is part one of my three-part series on building yours: brian-macdonald.com/guides/w…
1
54
Nobody really knows how to get more leads from ChatGPT yet. So companies are selling shortcuts. Some say they inject thousands of prompts into ChatGPT every night to push your business into its answers. That's black hat. I'm sure it'll get shut down, if it even works at all right now. Here's the white hat way. Clean, foundational work that builds a solid future and doesn't get you banned: • Publish one page for each question your customers really ask, with the answer in the first sentence • Add an llms.txt file: a plain text page that hands AI the fastest path to those answers • Add a footer link labeled "For AI agents" that points to it Google says it doesn't use llms.txt. We've seen the opposite. For one client, AI citations went up about 1,000% in the first two weeks. They started near zero, like most businesses do. Nobody's talking about this yet. Everyone will be. What goes in the file, with an example you can copy: brian-macdonald.com/articles…
2
91
Why do your AI agents keep giving different answers? Because each one is working from its own copy of your business. ChatGPT knows what you pasted in last week. Your inbox tool knows your emails. Your website chatbot knows the FAQ page from launch day. Ask all three the same question and you get three answers. The average company already runs 12 AI agents. Half of them work alone. And 68% of companies have traced a confidently wrong AI answer back to missing or inconsistent business information. A smarter model won't fix it. A smarter model just quotes the wrong price more convincingly. The fix is one place for the facts, a company brain, that every agent reads before it answers: • Each fact lives in one place • Every agent reads it first • Agents suggest updates, and a person says yes The agents are multiplying. The context isn't. Here's the full answer: brian-macdonald.com/articles…
3
2
99
How many leads are you getting from ChatGPT? In the first two weeks, one of our clients saw a 1,000% jump in AI citations. Last year 6% of consumers used AI to get a business recommendation; this year it's 45%. And AI doesn't hand them ten options, it names one or two. So the question is: when your customers ask about your products or services, does AI recommend you? I built an AI Search Visibility Report to answer exactly that: • 20 questions your customers actually ask, run through ChatGPT, Google AI Mode, Gemini and Perplexity: whether you're named, recommended or cited, and who shows up instead • What AI needs from your website to trust and recommend you, and whether yours has it • A ranked list of fixes and high value next steps, in plain words It's normally $200. For the first few people I know, it's $5. It runs through Upwork, so you'll need a free Upwork account. Here's the link: ownbox.io/AIVR You'll have your report and strategy plan about 1 day after I have your website address. Brian brian-macdonald.com/articles…
1
51
The most important database in your business is a person, and that person can quit. My AI Brain keeps what my business knows on my own Ownbox: type it, edit it, or add a photo and it reads the words. Try a working one at ownbox.io/demo
1
57
Your next customer is asking ChatGPT who to hire. It's naming somebody right now, and if it isn't naming you, that's the whole conversation. AI For Business OC is a monthly meetup for non-technical business owners and the people who run their operations. I open with a 15-minute talk on how AI search picks the businesses it recommends and what makes a local business one of them. Then we work, laptops open, on your actual website and your actual leads. You leave knowing where you stand in AI answers for your service and your city, and with the first fix already in motion. For Orange County local business owners and marketers. No technical background needed. Bring a laptop and a real question about where your leads come from. Location is shared with members. meetup.com/ai-for-business-o…
5
2
107
How many leads are you getting from ChatGPT? Ask it who to hire for what you do in your city, and read the names. Orange County, CA: free monthly meetup FOR NON-TECHNICAL ENTREPRENEURS in Irvine. DM me MEETUP for the invite.
57
Your company should be able to run a whole day without you touching it. Mine did today — twenty agents, three bosses, and one rule about who's allowed to ship. That rule is the entire thing. You don't need a beach, you need a chain of command that doesn't route through you. Comment TEAM and I'll send you mine, free. #claudecode #aiagents #buildinpublic #aiautomation
2
176
Every Claude Code teacher shows you the same thing: you, at the desk, in the loop all day. Watching your terminal from your phone is still you doing the work. Nobody teaches the company you build around it. That's the part that lets you leave. Respect to @austin.marchese @aiautomationsbyjack @robonuggets — start with them, then come back. Comment TEAM and I'll send you the chain of command, free. #claudecode #aiagents #buildinpublic #aiautomation
125
My best ideas never show up at the desk. I'll spend two hours stuck on a problem, paddle out, and halfway through a set it just untangles. My dad told me the same thing while he was building his tech company: step away from the problem, take a walk, and the answer will present itself. People think building something is about grinding harder — more hours, more screen, more coffee. It's the opposite. The work gets sharp when you step away from it, not when you force it. The paddle out isn't time off. It's the part where the hard problem actually gets solved.
1
1
131
Building a tech company stopped being the hard part. I can build faster than at any point in twenty years — and so can the guy copying me. My competitors raised tens of millions and hand every new customer $100 just to try their product. I can’t match that spend. I can be louder than them, for longer. Day one scoreboard: 100 local businesses handed their phones to Mavrick in month one. He answers, qualifies, books. The goal is 10,000 — and I’ll show you every step. The wins, the dead ends, the real numbers. Be loud. Start before it’s impressive. It’s the only edge money can’t buy.
1
1
113
I tried to trick an AI receptionist at a real law firm. New series. I pull up a real business on Google, Mavrick becomes their receptionist on the spot, and I call in trying to stump him. Episode one: a personal injury law firm. The move that matters most isn't what he answers — it's what he refuses. Ask about a service that isn't on their site and he says no and takes a message instead of bluffing. Ask if you have a case and what it's worth, and he gives zero legal advice — name, number, what happened, an attorney will call you back. That's the one thing a law firm's phone can never get wrong. Episode two: HVAC, and I'm calling at seven at night.
91
Twenty agents working at once have one real failure mode. It isn't bad code. It's that none of them know what the others just did. Two agents fix the same bug in the same hour. A third rewrites a file a fourth is mid-way through. Someone solves a problem that was already solved yesterday, and nobody finds out until the merge. This is what breaks multi-agent systems in practice — not capability, not model quality. The fix is one shared file every agent reads and writes. Here's how to build one, and the six rules that keep it useful past the first week. 1/6
1
115
Rule 5: Never write the wall from a snapshot. This is the rule I broke myself, and it's the one that silently destroys the channel. An agent that read the file an hour ago, appended its line to that old copy, and pushed has just deleted every post made in the meantime. The diff looks like one addition to the agent. To everyone else, it's a mass deletion. Two things fix it: always re-read the file immediately before appending, never from memory. And a check in the build that fails any change removing existing posts. The only correct change to a shared channel is pure addition. If your diff shows a removed line, stop. Rule 6: Put the writes through a lock. Two agents posting at the same moment will silently eat each other's messages. At twenty agents, this isn't rare — it's a Tuesday. One lock around the read-modify-write, held for the length of a file append, and the problem is gone permanently. A message that vanishes leaves no trace. Everyone believes it arrived. 5/6
1
16
Don't build all six today. Build the file and the injection hook — rules one and two — and use it for a day with whatever agents you already have. You'll discover the other four in the order I did, because each is something that breaks in a way you'll recognise the moment it happens: the chat log, the vanished message, the mass deletion, the post nobody could parse. Then come back and add the guard for the one that just bit you. One file. Injected, not remembered. Additions only. The commit history holds everything older than the newest posts — searchable in one command. Short and current in front, complete behind. The full breakdown, with starter prompts to build each rule, is in the piece linked in tweet 1. brian-macdonald.com/guides/h… 6/6
14