Loom vs Linear: A tale of two AI-cities
On building durable and sticky software in the age of AI and agents
“Because of rapidly advancing capabilities, agents will be brought into nearly all areas of work. Agents will be deployed to review every contract that gets written, handle the front lines of most customer support cases, audit every company’s financials, comb through every piece of medical research for drug discovery, generate nearly all code that ever gets written, create most sales and consulting presentations, transact across the web for consumers, and overall, be involved in nearly every other economically valuable task in society.” - Aaron Levie in “Building for trillions of agents”
In the old(!) world of SaaS, a “sticky” product was one that owned the interface. You built a beautiful, proprietary and walled garden, locked up the data, and you convinced users to live and work there. This is the basis of a system-of-record advantage.
But in the age of agents, what drives stickiness is changing. It’s no longer just about how much time I spend inside the UI, it’s about how easily your data can be pulled into my workflow. This is because, increasingly, peoples workflows exist outside of SaaS products, they live in agent harnesses like Gemini, Claude Code or OpenClaw.
It’s worth stating just how fundamentally novel (and interesting) this is. So much more of “the work” happens outside (on top of?) of SaaS tools here, it is done at a totally new layer of abstraction, which is agents.
To be sure, agents can (and will) interface with SaaS tools (databases, products, workflows, APIs), but they’ll never login to the product. They might not have a seat. They may not even know what the product looks like.
To say this changes a ton (everything?) about how software should be built and designed (and sold) is an understatement, but this is a point that many have made and been making for a few years. Instead of just making this point again, I want to use two products to explain what I see happening here, and how different approaches to AI/agent-readiness is playing out on the field. I’ll illustrate this point using two products I respect and enjoy using daily, Loom and Linear.
In case you’ve been living under a rock: Loom is a video messaging service for work. It helps with high-fidelity and async communication. It’s a beautiful, simple product. Atlassian acquired it for almost a billion dollars a few years ago. Linear is a project management tool with a particular focus on technology companies, product speed and craft. Linear is one of the most respected and loved products in all of tech.
TLDR: Loom is on track to lose in an AI-world because it treats its data as a destination. Linear stands to win because it treats its data as a primitive.
Loom is a brilliant product. It’s beautiful, the UX is brilliant, but it’s locked. It’s a walled garden. They want to sell me their AI, their summaries, their titles, their highlights. But what if I don’t want their AI tooling? What if I want to pipe a transcript into an LLM to extract action items directly into my own system? What if I want an OpenClaw agent to read through my Loom transcripts from a given week and post a weekly summary into Slack for my colleagues?
The answer is, I can’t. Loom (as of 10 March 2026) exposes no REST API to fetch transcripts or interact with my Loom content. I can’t use an API to fetch analytics, see who has watched or commented and I cannot post, edit or schedule anything programatically.
This is such a shame because Loom is such a treasure trove of rich, high-fidelity organisational context, exactly that which is scarce and that which makes agents powerful.
Linear, by contrast, is positioned to win in an age of AI, which is saying something, because project management tools are exactly the kind of thing that “vibe coding” makes fairly easy to replicate, because the primitives are so well-defined and known.
However, because they’ve prioritised a robust API and a very good CLI, they’ve essentially made themselves “agent-ready” which I think presents as a very viable and strong moat. “Be the best tool to plan, manage and complete work together with agents”. I suspect this is no accident and a result of clever strategy and product thinking.
Once you pair a seriously good API and CLI with agents, magic can occur. You can work at previously-impossible speed. Things that would take 30 minutes (audit and groom a backlog, remove stale issues, add new tags) now take seconds. You can add so much raw intelligence to your workflow. You can ask the model to ask you questions, to challenge you, to question your thinking. I can work on things in my Linear from my WhatsApp if I have OpenClaw hooked up. The best of all? If the agent doesn’t know how to do something, it will figure it out. It’ll use the “linear-cli --help” flag on the CLI to explore what’s possible with the API, and it’ll read docs if need be.
There was a joke that used to do the rounds online in the early days of AI that went something like: “AI is like having an intern, but it’s also like having an intern”, which was funny because it was true for the longest time, but I think we’re far beyond that now. It’s more like having a pretty damn capable additional set of eyes and hands by your side all the time.
In a world where more and more knowledge workers are using agents, building software products that are “agent-ready”, have a robust API, a killer CLI and an openness to that others may build and extend your product in ways you hadn’t planned, seems like the smartest play.
What does it mean to be “open”? It means embracing the fact that users will build things on top of you that you didn’t intend.
If I can’t retrieve a transcript via a command line interface, the software is a dead end to an agent. If I can’t programmatically upload a video, it’s a silo that an agent is locked out of. In a world of “vibe coding” and agentic workflows, silos are where 10x productivity goes to die.
Linear understands that they don’t need to build every feature anymore. They just need to be the most trusted, reliable and robust source of truth that an agent can talk to.
Products that force you into their specific “AI feature set” are fundamentally misunderstanding this shift. Winners in this shift may not build all the AI features, but they’ll be the most open and composable.
I love Loom’s core utility, but I’d be now open to move to something else. Not because the product is bad, but because I can’t use it where I increasingly want to work. I can’t plug it into my other tools and agents. I can’t compose new workflows and systems using its data and primitives. Ultimately, I can’t automate the tedious parts of the video messaging lifecycle.
Linear feels less and less like a piece of SaaS now, and more like a piece of agent-ready organisational infrastructure. On the other hand, I love Loom, and I hope they soon realise that an API is increasingly more valuable than a “Smart Summary ✨” button. If they don’t, they are building for a world that is rapidly changing before our eyes.