Your Startup Back Office Should Cost $150 Per Month

There is a version of me from three years ago who would have hired an office manager and a bookkeeper before he ever landed a client. That guy would have been out of money by March.

Instead, the operational side of Bridge AI runs on basically three tools, and for a brand new startup those three can be covered for $150 a month. No ops hire. No admin day. And the same setup is now running inside client companies in construction, marketing, real estate, accounting, higher ed, and IT.

The tools are not the interesting part. Anybody can buy them this afternoon. The interesting part is the setup. With less than two months of focused setup and learning, you can go from no back office at all to a fully functioning business. Almost nobody spends those two months, which is why almost nobody gets the payoff.

This week:

The three tools that run our whole company, and what those two months actually look like

How I built my own booking page in two hours instead of renting one

OpenAI cannot rule out critical cyber capability in its next model, Google reshuffles DeepMind, and a good place to steal other people's workflows

The Deep Dive

Your Startup Back Office Should Cost $150 Per Month

Here is the full list of what should run your startup back office.

Microsoft 365, about $30 a month for two people. That covers email, calendar, SharePoint for files, and Teams for meetings, so there is no separate video bill. Notion, $20. Claude, $100. That is $150 a month.

I can tell you Bridge AI runs on basically this exact list, with one main difference. We pay $200 for Claude instead of $100, because of how much we run through it, not just for ourselves but for client work.

Nothing on that list is impressive on its own. Plenty of businesses pay for all three and still drown in admin. The difference is Claude sitting in the middle, connected to every one of them.

Which brings me to the part people skip. How you set it up matters more than what you buy.

Start with the boring decision. Every kind of information in your business gets one home, and everyone knows where it is. Client records. Meeting notes. Hours. Files. Contracts. Pick the home for each one.

If that sounds like a lot of cleanup before you can begin, it is not. Claude will happily work with your mess. If a client is "Acme Construction" in your calendar and "acme" in your files and "Acme Co." in your notes, Claude can reconcile that, standardize it, and then keep it consistent going forward. It is not just working on top of the system. It is the back end.

So the reason most companies do not have this is not that their records were too messy. It is that nobody ever sat down and set Claude up. The tools sit there in separate tabs, unconnected, and everyone keeps moving information between them by hand.

Once Claude is actually wired in, the recurring work mostly stops being work. Calendar blocks become logged hours in Notion with no timesheet. Meeting transcripts land in a meeting notes database, tagged to the right client and the right people. A Friday scan of my sent mail finds every thread where I asked for something and never got an answer, with the nudge already drafted. Before a call, I get a brief with the email history, the calendar history, and research on the person and their company.

None of that needed a developer. It needed decisions, and then writing each workflow down once so it runs the same way every time.

Two months is the honest number. Month one is plumbing. Accounts, permissions, connectors, and the unglamorous work of agreeing where things live. Month two is building the handful of workflows you actually repeat. We have now done this for ourselves and for startup clients in various different industries, and it has been the same two months every time. Only the workflows change.

I will push back on my own point. This does not mean you never hire. It means what you hire for changes. You stop paying someone to keep records straight and start paying someone to do the work only a person can do. For a small company, that difference is the difference between being around in year two and not.

If you are starting something, or you are five people deep and the admin is eating your week, that setup is a big part of what we do.

From the Field

I Built My Own Calendly in Two Hours

I was about to sign up for a scheduling tool when I realized I would still be stuck with someone else's rules. So I built my own with Claude Code instead. Two hours from empty folder to live link, and I use it every day.

It is a Next.js app on Vercel with no database. Availability reads live from my Outlook calendar through Microsoft Graph, so there is nothing to sync. My working hours come from my Outlook settings, so when I change my hours there, the booking page follows. When someone books, it creates the event and the meeting link, and the guest picks Zoom or Teams.

The rules are why I built it. No same-day bookings, so the earliest anyone gets is my next working day. If someone lands on the page after hours or on a weekend, the following working day is blocked too. It counts back-to-back meetings before offering a slot, so if booking that time would give me a run of more than four in a row, the slot never appears. There is also a group link that only shows times when Tim and I are both free.

I did not even do the setup by hand. I had Claude drive my Chrome browser through the Microsoft and Zoom developer consoles while I worked on something else.

Every business rents a piece of software that almost fits. The gap between almost and actually used to cost a developer. Now it costs an afternoon.

AI News Worth Your Time

Three Stories From This Week

OpenAI says it cannot rule out "critical" cyber capability in its next model

On August 7, OpenAI published preliminary evaluations of Astra, an upcoming model, and said the results in agentic coding and cybersecurity were strong enough that it cannot rule out the Critical threshold under its Preparedness Framework. Critical means a model that can find and build working zero-day exploits in hardened real-world systems without a human, or run an end-to-end novel attack from nothing but a high-level goal. OpenAI paused internal work on the model that did not meet stricter security controls, added monitoring that can interrupt risky activity mid-task, and said it will test with government agencies and safety organizations. This is the second cyber disclosure the company has made in the past few weeks.

My take: this reads like a frontier lab problem, and it is not. The same capability that finds a zero-day for a defender finds it for an attacker, and the attacker does not need a security team to run it. Nothing exotic is required on your end. Multi-factor everywhere, patch on a schedule, and an actual list of who has access to what. That list is usually the scary part. READ THE OPENAI REPORT →

Google reshuffles DeepMind, and Jeff Dean leaves after 27 years

Sundar Pichai announced that Demis Hassabis is moving to Chair of Google DeepMind and Chief Scientist of Alphabet, stepping back from day-to-day operations to focus on AGI and science while continuing to lead Isomorphic Labs. Koray Kavukcuoglu steps up as SVP of Google DeepMind, running Gemini model development, frontier research, and the Gemini app and developer teams. Jeff Dean is leaving to launch an independent public benefit corporation with Sanjay Ghemawat, with Google as a founding investor and cloud partner. Pichai noted the Gemini app has passed 950 million monthly users, and Hassabis pointed to Gemini 4 as the next release.

My take: the org chart does not matter to you. What matters is that the biggest players keep sprinting, and the model you build on today will not be the best one in six months. Build your workflows so the model underneath is swappable. If your process only works because of one specific model from one specific vendor, you built a dependency, not a system. READ THE GOOGLE POST →

The Rundown launched a community for actual workflows

The Rundown, one of the bigger AI newsletters, opened a community hub where people post the real workflows they have built instead of talking about AI in the abstract. Screenshots, prompts, stacks, what broke. I posted my first one this week, the booking page from the section above, with the full build.

My take: the fastest way to get good at this is not a course. It is reading what somebody actually shipped last Tuesday and stealing the parts that apply to you. That is the whole reason From The Field exists in this newsletter, and it is nice to see a bigger version of it. Go look, and post one of your own. READ MY POST →

That's it for this week.

If this was useful, forward it to someone who is starting a company right now, or running one with a team small enough that they are still doing the admin themselves. That's who this is for.

And if you want to talk about what AI could do for your business, not in theory, but in your workflows, with your team, on your timeline, reach out at [email protected]. That's what we're here for.