Should an agency build its own super app?

Rarely. Building one takes engineers on staff for good, because every tool needs sharpening each month. Space, the AI-native brand agency, built Agency OS for its 107-person collective. It's a super app that replaced 28 apps we used to pay for, on one database with one login. Space now sells it to other agencies, so they can skip the build.

What is Agency OS?

Agency OS is the operating system Space runs on, sold to other agencies. It's one app with one login. Time tracking, invoices, the CRM, docs, contracts, video, scheduling, and dictation all live inside it.

Across its 14 categories, it does the work of the 42 most popular standalone apps agencies pay for today.

Why build software when you can rent it?

Space has one rule for tools. It replaces an app only when its own build is far better than the best product it could buy. Space Labs is the public shelf of the 21 tools we built under that rule. Recon is the cleanest case; competitor tracking tools run $15,000 to $20,000 a month per brand. Recon costs a few dollars. (The full list of what those tools changed is in trade-offs we broke.)

Agency OS turns the rule inward. A rented tool comes with someone else's roadmap. It comes with their bloat too, because it has to cover every edge case for every customer. Agency OS was custom made inside an agency, for how agencies work. Every tool in it reads from a single source of truth. The companies behind the 28 apps Space replaced are worth more than $76 billion combined.

What changes when every tool shares one database?

A client in the CRM is the same client in your time entries and your video reviews. Nobody copies data between apps, so nothing drifts out of sync.

Provisioning happens once. A new crew member gets one login on day one. When someone rolls off, we flip one switch. (At 107 people, per-seat pricing had become a real line item.)

At list price, the top self-serve plans of all 28 apps would cost Space $1,383,249 a year. That's $12,928 per person. Those savings go to clients. Space spends less on software than an agency renting 28 apps, so it can lower its prices without cutting senior time. An agency on Agency OS can do the same for its own clients.

The tools also trigger each other, with no Zapier and no glue code that breaks on a Tuesday. Copy the scheduling assistant on any email thread. It reads your calendar and sends the guest a few open times. Once they pick one, it books the meeting with a Google Meet link. If it needs something from you, it asks in Slack.

You learn the interface once. The three video tools look and work the same, so knowing one means knowing all three. Fewer context switches mean more use; when Space turned its writing skills into one-word Slack commands, team usage jumped an order of magnitude.

Agency OS also connects to Claude and ChatGPT over MCP. That means you can search or change anything inside it with a plain prompt. Ask for a doc or a spreadsheet; it shows up in Agency OS. Ask for a change across the whole database; it happens in one pass. A few prompts that work today:

• "Rename this client everywhere it appears in our workspace."
• "Find every time this client praised our work in a call, email, Slack message, or comment. Pull the exact quotes for their case study."
• "Read this week's call transcripts against our evergreen notes on this account. What changed?"
• "How many hours did the team log on this client in September, by person?"

Both assistants can also pair Agency OS with your other connectors. With Gmail and Slack connected, one prompt can search your inbox, Slack threads, call transcripts, and doc comments together. Then it logs what it found on the client's record. (The search reaches inside the recordings themselves, down to the sentence someone said.)

Why is a super app usually a bad idea?

For a mass-market product, a super app is a terrible idea. You'd build fourteen average tools for millions of people who all want different things. Building for one kind of company flips that math.

It helps that Space is the user. We're Agency OS's first customer, so there was no one to research. We already knew which features we wished existed and where the old tools fell short.

Agency OS has none of the bloat of the tools it replaced. It also has things those tools never had, because they fit how agencies work.

• Notetaker records Slack huddles, plus Zoom, Google Meet, and Microsoft Teams calls. It runs in the background with no bot in the meeting. Video and a transcript land in a couple of minutes (other recorders take 5 to 15).
• A dictation app loads four kinds of words into its dictionary: your team's lingo, each client's brand name, their product names, and the names of the people on their team. Every transcript draws on all past calls and every correction anyone made. The system learns from its own mistakes, so you stop fixing the same error over and over. I use it 50 to 100 times a day.
• An email ghostwriter reads each person's entire email history. It learns how they react to different kinds of emails in different situations. Then it drafts Gmail replies in the tone that person would use for that one. It learns on its own, for every person on the team. (It never hits send.)
• A monthly pulse check asks the team how happy they are out of 5. Anything below a 5 gets a follow-up asking what would get them there.

Some of these tools had nothing to replace. Space looked, found no product that did the job, and built it from scratch.

A super app is bad software for a million people. For 107 people, it's the best software I've used.

I think it goes further, down to one person. More on that in will everyone have their own software?

How does one agency run smarter AI than a SaaS company?

Bloat is half of why standalone tools fall short. Every edge case becomes a feature; the product gets worse with each release. The other half is cheap AI. A mass-market SaaS company has to protect its margins across every kind of customer, so it picks the cheapest model that passes. Few run the smartest models available.

Space picks the model behind every tool, quality first. Agency OS runs its agents through OpenRouter, so it can reach past Claude and ChatGPT to any lab's models. Each task goes to the model that does it best for the least money. Agency OS skips the bloat, so the budget goes to smarter models. It still costs less.

Dictation shows the gap. One popular dictation app lets you add as many dictionary entries as you want, but it only reads the first 50. Ours reads thousands. It runs on a smarter model, about three times faster, at a tenth of the cost. It was also more accurate than every incumbent we tested.

Notetaker and the ghostwriter follow the same pattern. Each runs on a higher-end model than the tool it replaced. Each came out faster and cheaper in our tests. The output was better too.

What's the catch?

Lotus Notes tried the one-app idea in 1989. Its pieces fell behind the standalone tools they competed with, so users drifted off one app at a time. That's why Space's rule sets the bar at far better.

Upkeep is the other cost. Models improve every month. A tool that wins today goes average within a year if nobody sharpens it. Space does that sharpening for every agency that runs Agency OS.

When should an agency build instead of buy?

Almost never from scratch. My view, by size:

Who is Agency OS the wrong call for?

Agencies that work inside their clients' tools. Some enterprise clients can't approve a third-party app, so the work has to live in their stack. It's also the wrong call for small shops (under about 20 people), where the per-seat savings won't cover the switch. It's also the wrong call for slow legacy agencies locked into long, expensive contracts with their current software.

More from Space: inside Space Labs and we sold you the PDF.

Run an agency and want to see it running? Say hi anytime.