Macro - Unifies your work into one app with shared memory

@linked  @jacob_beckerman The 'shared memory' angle is clever — most tools treat memory as siloed. What made you prioritize cross-app memory over just being another workspace app?

@linked  @jacob_beckerman The detail that stood out to me in the launch notes: CRDTs instead of last-write-wins for docs. Most teams never get that far, so the failure class is clearly on your radar. The part I'm curious about is the unified brain. Docs are mergeable, so CRDT closes them cleanly. Team-level memory looks different: a background refresh and a live agent write can land on the same memory object, and those writes don't merge, one of them just wins. Same question for an agent that read a memory value a few steps before acting on it, after the refresh already moved it. Does the memory plane get the same write-time discipline as the docs, or is that handled somewhere else? Genuinely curious, the docs choice suggests you've thought about it.

@1mirul Thanks! Yes, we think it is a pretty obvious pitch. When we set out to build it I knew I wanted it, but wasn't sure how good it would all feel. Now two years later with much of it polished, I can say it feels really really good. Our company is decluttered, everyone is aligned and we're shipping fast. It's not rocket science but it is super tedious design work. I take pleasure in doing it better than anyone else would have done it, if anyone else would have tried. The existing stuff is good, it's just not great and it's not well-integrated for teams and agents, and it's closed-source. Thanks!

@monir_ You can also accomplish this by MCP'ing Claude/Chat into all your tools. But:

• MCP is often a limited subset of the app itself, or rate limited. Your agent received a partial representation of what you get as a human user. It often makes mistakes that aren't forgivable. Incentives are misaligned because each develop wants to keep you in their ecosystem, just just become something that lives under the LLM.

• You're still paying for all of those tools. The cost adds up to much more than Macro's free version or $40/month plan. if you sum Superhuman, Notion, Linear, Slack, HubSpot, etc., etc., you get to a significant sum. And to save money, often you'll not give e.g. HubSpot seats to everyone who needs them which leads to further fragmentation of who-can-see-what in your company. Macro unifies all of this into one business system.

• You still have to login, 2FA and have sign ins to a bunch of different tools. This is mostly a problem on mobile where you have links going across apps. It's easier, but still annoying, to context switch between different tabs on desktop. It's even harder on mobile.

• Fundamentally, each tool was designed as a silo, only with limited integrations. For example there is no button in Superhuman to take a customer email and report it as a Linear ticket: that's a manual process you have to do of pasting an image in Slack and having an engineer (or you) create a ticket, which doesn't always happen. In Macro, because each product surface was co-designed you don't have to do this, just hit "task" from email to bidirectionally link a new task to that customer email.

• There is no unified memory layer, in part because nobody is "in charge" of constructing this. Nobody has all the pieces. You buy a little slice of your stack from each vendor. Macro provides unified and team-level memory across all your tools, because it is all your tools, co-located in one ta and co-designed to work together.

@vikramp7470 Thanks Vikram - we believe the successor to legacy SaaS will be an open, modular, extensible workspace. I also believe most startups die from apathy: the most important thing you have to do as a founder is make a dent in the universe. If we succeed in improving things for users, I'm confident this is a great business as well. And over time as Macro matures, it will be silly to use proprietary closed-source SaaS when there's a unified and open alternative.

Right now there's a mobile app that's native with Tauri, which we will also use to make the Mac and Windows apps eventually. For now it's browser-only on desktop - would this be a top priority for you? Personally I prefer email in a desktop app, so I see the use caee, but since Arc browser I'm mostly using we versions of apps, except sometimes Figma for local fonts.

@zehra6lwq Recall (in the sense of precision and recall) is a problem for all AI systems, and humans too ;). One of the biggest benefits of having all your workspace in one database (Macro) not spread across a bunch of apps is the unified memory aiding in recall. Macro's Unified Search tool available to the agent allows it to search through all content types, rather than having to execute multiple MCP calls, dedup resonses, order chronologically, then make more calls to further investigate. Of course, our approach saves cost (tokens) and time (tokens, again) but it also improves performance on these types of queries.

@dipankar_sarkar Yea, there's no right answer and it's a hard problem. From a design perspective there's a couple options (i) let some LLM decide (ii) manually grant permissions at runtime (iii) the agent inherits permissions of some human. Four our case, at least for now, we've chosen ~3:the access control inherits from the user that launched the agent or that referenced them in the channel. This works pretty well and hasn't let to any embarrassments yet, since your agent is restricted to accessg things you have access to, it generally matches users' expectations and fails understandably.

@tobiasfleischer What is worth remembering? That's a good research topic! Probably, whatever is salient, especially in light of future requests you expect the user to make. For us, right now, it depends highly on what you connect and who you're working with. My memory system is mostly filled with biographical details, history of the company (Macro), and what things I prefer and don't in AI responses, who I work with and what they're the technical owners for, and some quite personal information about my partner and pets (picked up from my connected personal email) and family.

(if you want to know what Macro knows about you so far, you can ask for it's memory dumb - after using it for a few months mine is super impressive)

@novamaker01 Great question. We have what we call Channel based permissions permissions (see https://docs.macro.com/product/channels) which means whenever you something in a channel is is auto-shared with participants. Compare that to, say, Notion + Slack or GDocs + Slack, where you need to also remember to manually share your link with everyone in the channel. Then, the memory system is built from your permissions because all Macro agents inherit your permissions; so memory includes all channels you're a part of, all email accounts connected, external services via MCP, and the text content of all channels, including Macro Calls aka standups/huddles spawned from those channels!

As I'm writing this, I'm realizing how important permissions is for building team-level memory. TBH we don't think about this much internally as we use Macro because it "just works". But yes, it would be a lot harder otherwise because you have no centralized permissions controller.

@thenameisarian Thanks for your comments Mustafa - see my reply to Dipankar!

@omri_ben_shoham1 I agree that historically you've had to choose between Let's run trough some comps of who's tried and why I think it went wrong. It's important to note that all of these attempts were pre-coding agents, so a reasonable answer to "why now" is "this problem [rebuilding the workspace] literally just became tractable". Of course, we started building Macro before coding agents, but they've allowed us to make progress on our roadmap much faster and expand our ambitious to more blocks.

• Notion chose the wrong level of abstraction. Markdown docs and databases are super flexible but they're not as as good as purpose-built products. We ran our last company on it but had to migrate CRM and Tasks off of it as we scaled, leaving Notion hollowed out. Also, Notion really isn't an all-in-one, it's docs/notes/wikis and now agents. They just unshipped Notion Mail. They don't have calls, channels or mail. Their calendar is a different app that they acquired. So I think it's fair to say they're backing off the "all-in-one" pitch to focus on agents.

• ClickUp figured out that most F500 teams can't discern slop from quality, and that sales matters more than product for that market. So they rapidly built an okay product with that market and layered sales and marketing. They never won with taste-makers and startups. They're super closed source, and not even as friendly as Notion when it comes to dev extensibility. But they were the only game in town for the all-in-one pitch. But still, to my knowledge, no email client integrated? That's an important part because it's where customer requests come in and it's the main comms channel for sales, customer support, etc.

• Coda was like Notion but expanded a bit too quickly IMO whereas Notion stayed focused on docs, wikis and tasks for longer. Coda never won the mindshare that Notion got with solo users (and then lost to Obsidian a few years later, as they enshitified and moved upmarket).

• Quip was a great mid-2010s attempt at this from Brett Taylor. I've asked a few people why this didn't go as planned and I'm not so sure why. They did spreadsheets, unlike Notion, and channels and it was pretty sleek. It got adopted by some big companies. Perhaps it was mismanaged by SFDC, I'm not sure. It had a very different vibe than Macro IIRC, it wasn't as good as the standalone tools, it was more of a light version of docs, spreadsheets, etc.

TL;DR: Macro is the only tool that doesn't make you choose. You get (i) best-in-class blocks (notes, email, etc.) as well as full integration across the suite. Coding agents are what make building this feasible.

See https://macro.com/posts/ for some more info about how we've designed Macro and how it compares to what came before it, and how those tools inspired/informed us!