Unbundling SaaS: How to Expose Your Product's Capabilities to Agentic Systems
What does it mean to unbundle a SaaS for agents?
A SaaS, at its core, is a UI on top of a database for people to perform stuff themselves .
Unbundling means taking the capabilities that used to live behind that UI and making them directly callable by software that can reason and take action.
In practice, that means decomposing the product into discrete capabilities an agent can invoke — and shifting the UI from a place where humans do the work to one where humans (if still in the loop) review it.
Old shape: human logs in, clicks through screens, performs the task.
New shape: an agent calls the capability; a human reviews the proposition.
How do I make my SaaS agent-ready?
Start by listing the capabilities an agentic system would actually need from your product, then expose them in the right way.
For a sales engagement product like Lemlist, that looks like: Lists of leads Enrichment — finding emails Warm-up domains Sending emails and follow-ups Triggering calls The underlying bet is that small, one-person or prosumer software is highly likely to be replaced by a very good agentic loop with a chat connected to the right capabilities .
Making your product one of those capabilities is how you stay in the loop.
“a SaaS is, it's basically a UI on top of a database for people to be able to perform stuff themselves”
Should I build an MCP server for my product?
The broader pattern is a massive wave around Claude Code and its multiplayer equivalent ("the version for everyone of cloud co-work") where users bring their plugins and connect agents to capabilities.
If you want your product to be one of those plugged-in capabilities, you need a deliberate surface for it.
“most agents' initiatives just get stuck at the planning phase or the integration phase because the agent isn't really able to do the work that a person wants it to do”
Why do most agent initiatives stall on today's SaaS?
Agents are only as viable as their ability to act in the real world, and most business work still happens through a person clicking around a browser.
That is exactly where things break down: Most agents' initiatives just get stuck at the planning phase or the integration phase because the agent isn't really able to do the work that a person wants it to do for him.
Browser use and computer use are stopgaps for SaaS that hasn't been unbundled — they let an agent drive the human UI when there is no clean capability surface to call.
“we have a a dev with who is now almost full-time working on the cloud code setup”
What does unbundling look like inside the vendor?
This is not a side project.
At Lemlist, adapting an established product to an agentic interface required dedicated engineering: A dev almost full-time working on the cloud code setup .
An entire file system of skills, rules, subagents so the agent is actually helpful and efficient in the codebase.
An acknowledged learning curve, especially on a codebase with nearly ten years of history.
Translation: exposing capabilities to agents is itself a product workstream, not a config change.
Frequently asked questions.
- What is unbundling a SaaS in the context of AI agents?
- It's the move from a UI-on-top-of-a-database that humans operate, to a set of capabilities an agentic system can call directly. The Lemlist example is decomposing the product into things like lists of leads, email enrichment, warm-up domains, sending emails and follow-ups, and triggering calls — each exposed so any agent that needs them can plug in, with a human optionally reviewing instead of doing.
- How do I make my SaaS agent-ready?
- Inventory the capabilities an agent would need from your product and expose them in the right way for agentic systems to consume. The underlying shift is from features designed for humans to perform work, to features designed for humans to review work an agent proposed or executed. Plan for the engineering: at least one product team treats the agentic setup — skills, rules, subagents — as a near-full-time workstream.
- Should I build an MCP-style plugin or capability surface for my product?
- If you believe small, prosumer, and one-person software is likely to be replaced by a good agentic loop with a chat connected to the right capabilities, then yes — being one of those callable capabilities is how you stay in that loop. The wave around Claude Code and its multiplayer equivalent is explicitly built on users bringing their plugins.
- Why do so many agent projects stall today?
- Because most business work still happens through a person on a browser, and most agent initiatives get stuck at the planning or integration phase since the agent can't actually do the work a human would do. Browser-use and computer-use approaches exist precisely to bridge SaaS that has not yet exposed its capabilities to agents directly.
- Is exposing capabilities to agents a small config change?
- In a mature codebase it is a dedicated workstream. Lemlist has a developer almost full-time on the Claude Code setup, building out a file system of skills, rules, and subagents so the agent behaves usefully inside a codebase with close to a decade of history. Expect a learning curve before the agent is efficient in your environment.