SaaS Still Has a Moat. Just Not the One We Think.

When I consult with companies trying to become AI-first through Second Coffee, I keep seeing the same pattern.

They want agents. They want a company brain that knows everything without anyone copying and pasting. They want their data connected across the whole stack.

But they are not ready to get rid of their SaaS.

That observation has changed how I think about SaaS moats and whether they are going away.

For years, everyone assumed the moat was data.

Salesforce owned your customer data, sales history, pipeline records, reporting structures, permissions, automations, and years of institutional memory. Getting that data out was painful. Switching meant migration risk, broken integrations, months of retraining, and a lot of organizational pain.

Same for HubSpot, Workday, Zendesk, Airtable, SharePoint, Notion, and the rest of the stack. Each one sat on top of a pile of data your company couldn’t easily move. The data was the lock-in.

That moat is weakening. And something more interesting is taking its place.

The Real Moat Is the Habit

The data moat was real. But it was never the only thing keeping companies from switching.

SaaS’s true moat is the daily habit of using the UX.

Millions of salespeople have been trained to use Salesforce daily. Teams have built dashboards, custom fields, permission structures, and reporting workflows inside it that they use frequently. Managers know where to look. Sales reps know what to update. Integrators have spent years perfecting the Salesforce integrations.

And it’s not just Salesforce. Ops teams know which Airtable view matters for Monday morning. Customer Success teams have been trained and optimized to use ZenDesk. Each of these people are so familiar with their apps’ UI that they no longer even think when using them. Learning a new tool would force them to start all over.

Even more - once SaaS has been integrated with custom workflows, it becomes a matter of pride. Small teams building on top of Airtable or Notion often have something close to the IKEA effect: they built the workflow themselves, so they trust it and take pride in it. Even an imperfect interface becomes comfortable because it structures how work actually gets done.

Every department wants its SaaS to be the system of record. CS wants the CS platform. Engineering wants the product database. Sales wants the CRM. Marketing wants the marketing SaaS. No one voluntarily hands their system over to a replacement they didn’t choose.

A new AI-native tool may have a genuinely better interface -- more agentic, with dashboards that build themselves, a chat interface that can accurately answer any question about the business, and automations that do half the work before anyone asks. But adopting it means retraining employees, rebuilding dashboards, recreating permissions, migrating custom fields, and convincing every team the pain is worth it.

People complain about Salesforce constantly. But what they usually want isn’t a replacement. They want Salesforce to be more powerful with more data and also be less annoying to use. They will therefore not give up Salesforce without a huge amount of friction and feet-dragging.

People in the company who have optimized their own workflows based on the SaaS UI don’t want it to. And honestly management would not be happy either as in the short and medium turn, it would likely cause a reduction in productivity instead of an increase.

Big SaaS has a strong foothold right now, and it will last longer than people think.

While safe for now, the future does not look bright for SaaS

While SaaS has a strong foothold with UX and habit, real headwinds are coming that will limit its growth and eventually cause a reshaping of the industry.

They will not dent SaaS revenue right now, but they will materially impact it in the coming years. Here are five headwinds that will diminish their dominance.

1. AI is forcing data to become portable. Companies are starting to deploy AI to help with sales, support, marketing, and data analysis. To be successful, this AI needs context across the whole stack -- not just one system. That’s creating pressure for SaaS companies to build out APIs, CLIs, and MCPs -- ways for agents to access and manipulate their data without needing a human in the loop. SaaS vendors that don’t open up will get routed around. SaaS vendors that do open up make their data easier to extract -- drastically lowering the data moat.

2. Agent-to-agent makes human UX less relevant. As more work gets done by AI agents talking to other AI agents, the beautiful human interface stops being the point. If my agent is querying your system via API, your UI doesn’t matter. What matters is your data model and your API surface. That’s a different competition entirely -- one where the incumbents’ biggest investment (the interface and user habit) becomes less of an advantage.

3. New data paradigms make old structures a liability. SaaS was built on relational databases. AI is increasingly built on context graphs, vector stores, and tree-structured data in markdown and JSON (formats that relational tables can support but are not optimized for). As AI-native data architectures become the standard, the existing data structures that once created lock-in may start to look like technical debt.

4. Internal context layers may make individual SaaS tools less necessary. Companies building AI-first infrastructure are starting to pull data from across their stack into a unified context layer. Once that layer exists and is well-maintained, the marginal value of any individual SaaS tool goes down. The context layer becomes the system of record. The SaaS tools become input sources. This is happening most frequently with new, fast-growing startups -- which was one of SaaS’s major sources of future growth. SaaS is losing its farm team.

5. AI coding is flooding the market with alternatives. Building software used to be hard. It isn’t anymore, at least not relatively. The same dynamics that let a small team build a sophisticated product in weeks also means there are now dozens of cheaper alternatives to every incumbent. None of them are as good. But they don’t have to be. They just have to be good enough at a fraction of the price, and they will apply constant downward pressure on pricing and perceived value.

None of these will completely dissolve Salesforce next year or even in five years. But all five are encroaching on the edges, weakening their stronghold.

What To Do About It

If you are a large SaaS company

Your data moat is weakening whether you like it or not. The right response is not to resist it -- it’s to get ahead of it.

Build out your MCP and API surface aggressively. If your data is going to become accessible, you want to control how and earn relevance in the agent ecosystem rather than get bypassed by it.

Own the context orchestration layer. Don’t just be a data source. Become the system that connects data across tools, manages permissions, and routes context to agents.

Invest in a new AI-native data backend. Buy or build a contextual database that can handle graph structures, vector retrieval, and the data formats AI actually wants to work with.

Make it easy (and cheap) to have companies’ processes run and grow on your platform. Bundle it in even if you take a small loss initially. That’s the new moat and the ones who execute will stay in the game.

If you are a company tired of SaaS sprawl

I have been instructing companies I work with not to rip anything out immediately and replace it with AI. People hate change and you will see immediate resistance.

Instead, start by building things that the existing platforms can’t do. Rather than replacing your CRM, stand up a context layer that syncs with your existing tools. Pull data from Salesforce, HubSpot, Airtable, SharePoint, and wherever else it lives. Don’t replace those tools -- just link them.

Then start building agentic workflows on top of that layer. Bring in new data sources your current SaaS tools don’t support. Build new interfaces for capabilities your teams don’t have today -- things that weren’t possible before all this data was connected. Cross-functional search. Automated briefings. Agents that know the full context of a customer relationship across every system.

Give your teams abilities they couldn’t have with any single SaaS product in isolation. Let them build on the new layer. Let them get comfortable with it.

And then, slowly, start pulling the plug on each SaaS tool. Not all at once. One at a time, when the context layer has fully replaced its value. Do it team by team, starting with the ones with the weakest habit moat and the most to gain.

The incumbents are counting on the switching cost to protect them. You can make the switch gradual enough that it barely feels like a switch at all.

If you are a startup competing with SaaS

If you are a startup competing with Big SaaS - Don’t try to replace the incumbents head-on. You will lose the habit war.

Instead, own the new context layer and sync your data with the existing players. Make it easy to pull from Salesforce, HubSpot, and the rest. Don’t ask anyone to abandon what they already use.

Then build interfaces and capabilities that don’t exist yet -- things the incumbents can’t easily add because they’re constrained by their existing data model and their existing user expectations. A unified company memory. Cross-stack search. Agentic workflows that span multiple tools.

As agent-to-agent communication becomes the norm, the power structures shift. The incumbents’ advantages -- huge installed user base, familiar UI, years of customization -- become less relevant when the end user is another AI. Your advantages -- flexible data architecture, clean API surface, native agentic design -- become more valuable. Position for that world now, even if it’s a few years away.

The Timeline

SaaS isn’t dying next year or even the next three years.

But the five pressures above are compounding. And the companies -- both startups and enterprises -- that start building now for the world where agent-to-agent is the norm, where context layers replace SaaS as the system of record, and where AI-native data structures have replaced relational tables, will be the ones who end up on the right side of this.

Data is easier to move than habit. But habits can be replaced. It just takes longer than a demo.

(Original posted at https://iamcharliegraham.substack.com/p/saas-still-has-a-moat-just-not-the)