I was in a board meeting last week for a breakout AI B2B leader at $100m+ ARR and the team was excited … but stressed.
Why?
So many of their largest customers were happy to sign relatively large checks for their AI Agent. Six figures. Some even seven. The product was working. Customers loved it. Pipeline was strong. The team had every right to be proud.
But here’s what kept coming up in almost every deal: “We’ll sign. For one year. To start.”
Not “we’ll sign a three-year deal with a ramp.” Not “let’s lock in pricing for the long term.” Just … one year. Because AI is changing so fast, they’ll see about next year.
So churn risk was essentially embedded in every single deal. Even the big ones. Especially the big ones.
I get it. As a buyer, I’d do the same thing right now.
The New B2B AI Reality: Everyone’s Happy … For Now
Let me be clear about something: this startup I talked about above is crushing it. Growing fast, delivering real value, customers are genuinely delighted with the product. They may well have the best AI Agent in their space. This isn’t a story about a struggling startup. This is a story about a breakout AI leader wrestling with a structural problem that’s going to hit every AI agent vendor in the next 12-24 months.
The problem isn’t product-market fit. The problem is that the switching costs in AI agents, at least in some cases, are structurally lower than what we’ve seen in SaaS before.
In traditional SaaS, switching was brutal. Migrating your CRM meant months of data migration, retraining hundreds of reps, rebuilding integrations, reconfiguring workflows. The pain of switching was so high that even mediocre vendors could maintain 90%+ gross retention for years. That switching cost was the moat. It was the foundation of every SaaS valuation model ever built.
AI agents are different. Fundamentally different.
Prompts Are Portable. And That Really Does Change Everything.
Here’s what really brought this home for me. We recently added a new AI sales agent at SaaStr. And to train it, we started by taking a prompt from another vendor that worked … and just giving it to the new vendor.
Yes, we still had to train it. There was fine-tuning, testing, iteration. That part is real.
But the prompt was portable. Maybe 50-80% of the migration work was done just by cutting and pasting a prompt in a few minutes.
In old SaaS, switching your marketing automation platform was a six-month project with a dedicated team. We’re tried to leave Marketo for 5+ years. We can’t. But switching your AI sales agent? You copy-paste the core prompt, spend a few days tuning it, and you’re largely up and running.
We’ve seen this firsthand across our 20+ AI agents at SaaStr. The UI looks different vendor to vendor. The terminology varies — some call it “coaching,” others “instructions,” others “system prompts.” But it’s fundamentally the same process. Which means once you learn one agent deeply, you can deploy a competitor’s agent frighteningly fast.
We actually maintain a library of our best prompts, tone guides, and training materials now. Not because we’re planning to churn. But because we know we can. And that leverage changes every vendor conversation we have.
What This Means for AI Agent Vendors
If you’re building an AI agent company, this should be sobering. Let me be direct:
Your AI isn’t your moat. Your AI is table stakes.
When we deployed an enterprise AI agent from Salesforce using copy-pasted prompts from a two-year-old startup — and both worked great — that tells you something important about where defensibility actually lives in this market. (Hint: it’s not in the model.)
So what is the moat? A few things still matter:
- Integration depth. Native Salesforce integration is genuinely better than bolted-on. Deep workflow embedding — where the agent is woven into the customer’s actual operating rhythm — creates real friction. Not because the AI is better, but because the plumbing is harder to replicate.
- Specialized infrastructure. Email deliverability, domain reputation, compliance frameworks. These are boring. They’re also actually defensible. A customer who’s been building sender reputation through your platform for 18 months has a real reason to stay.
- Network effects. Does your agent get smarter from aggregate customer data? If your product improves as your customer base grows, you have something. If every customer’s agent is essentially independent, you’re just a wrapper around an LLM.
- Vertical depth. A medical coding agent or legal contract review agent with years of specialized training and proprietary datasets? That’s genuinely hard to replicate. A generic GTM agent? The moats are paper-thin.
What’s notably absent from the moat list? The actual AI quality. That’s the uncomfortable truth of 2026.
The One-Year Contract Problem
Let’s get back to that board meeting, because the one-year contract dynamic deserves more attention.
In traditional SaaS at $100M+ ARR, you’d expect to see a healthy mix of multi-year deals, especially in enterprise. Three-year contracts with annual escalators were the gold standard. They provided revenue predictability, reduced churn mechanically, and gave you time to prove value and expand.
In AI agents, buyers are explicitly refusing to commit beyond one year. And it’s not because they’re unhappy — it’s because they’re rational.
Think about it from the buyer’s perspective:
- The underlying models are improving every quarter. What’s state-of-the-art today might be commoditized in six months.
- New entrants are launching weekly with increasingly capable offerings.
- And as we just discussed, the switching costs are dramatically lower than traditional SaaS.
So why would any smart buyer lock in for three years? You’d have to offer a massive discount to compensate for the optionality they’re giving up. And even then, many won’t do it.
This creates a structural challenge for AI agent companies that I don’t think the market has fully priced in yet. When every contract is effectively a one-year deal, your gross retention rate becomes your actual retention rate. There’s no mechanical multi-year buffer smoothing things out. Every customer is making a fresh buy/no-buy decision every twelve months.
That means if you’re an AI agent company projecting 95% gross retention, you’d better be sure. Because in a world of portable prompts and one-year contracts, 85% gross retention — or even 75% — is entirely plausible for vendors that don’t build real stickiness beyond the AI itself.
The Math Gets Scary Fast
Let’s run some rough numbers on what this means at scale.
A traditional SaaS company at $100M ARR with 92% gross retention loses ~$8M per year to churn. Painful but manageable, especially if you’re growing 40%+ and expanding within accounts.
Now imagine an AI agent company at $100M ARR where gross retention drops to 82% because of prompt portability and one-year contracts. That’s $18M in annual churn — more than double. To maintain the same net growth rate, you need to sell an additional $10M per year just to stay even. That’s $10M in extra bookings that goes straight to replacing churned revenue instead of driving growth.
And here’s the really scary part: in a world where the technology is improving this fast, the customers who churn aren’t going back to doing things manually. They’re switching to a competitor. So your churn is your competitor’s new business. It’s a zero-sum knife fight for the same budget, quarter after quarter.
What Smart AI Agent Companies Should Do About It
If I’m advising an AI agent company right now — and I’m on the boards of several — here’s what I’d be thinking about:
1. Build stickiness beyond the prompt. Your prompt layer is portable. Accept that. Now build everything around it that isn’t. Proprietary data flywheels. Workflow integrations that take months to set up. Compliance infrastructure. Make the 20% of migration that isn’t just copy-pasting a prompt as painful as possible.
2. Go vertical, fast. Generic horizontal agents will face the most churn pressure because they’re the most substitutable. A purpose-built agent for insurance claims processing with deep regulatory knowledge and industry-specific training data? That’s a much harder thing to replicate with a copy-pasted prompt.
3. Rethink your pricing model. If customers won’t commit to multi-year terms, maybe the answer isn’t to fight it. Maybe it’s to lean into consumption-based or outcome-based pricing that aligns your revenue with the value you’re actually delivering. If you’re billing based on results — meetings booked, tickets resolved, claims processed — the prompt portability matters less because you’re competing on outcomes, not features.
4. Invest in customer success like your life depends on it. Because it does. When every customer is making a fresh decision every year, the relationship between renewals and your CS investment becomes almost 1:1. The companies that underinvest in CS because “the product sells itself” are going to learn an expensive lesson.
5. Ship faster than prompts can replicate. The best defense against prompt portability is to keep innovating at a pace where your latest capabilities haven’t been reduced to a transferable prompt yet. If you’re shipping new features monthly that deliver genuine incremental value, you’re creating a moving target that’s harder to copy-paste away from.
What Smart Buyers Should Do
If you’re a buyer of AI agents, this dynamic should be liberating. You have more leverage than you’ve ever had purchasing B2B software. Use it.
Negotiate hard on contract terms. One-year deals with reasonable exit clauses should be your default. Any vendor pushing hard for multi-year commitments in this market is telling you something about their confidence in their own retention.
Keep your prompts organized. We maintain a library of every prompt, training document, and configuration file across our agent stack. Not because we’re planning to switch everything tomorrow, but because the option to switch efficiently is itself a form of leverage.
Run bake-offs regularly. The cost of testing a competitor has never been lower. Take your best-performing prompt, hand it to an alternative vendor, and see what happens. You might be surprised. Or you might confirm that your current vendor really is the best. Either way, you’ll negotiate your renewal from a position of knowledge rather than inertia.
Don’t consolidate too fast. The temptation to put all your agents on one platform is real. But in a market moving this quickly, maintaining optionality across a few best-in-class vendors per use case is often smarter than going all-in on a single platform.
Latent Churn is There in AI Agents. You Just Can’t See It Yet.
The AI agent market is going to see a wave of churn unlike anything B2B SaaS has experienced before. Not because the products are bad — many of them are genuinely great. But because the structural switching costs that propped up SaaS retention for two decades simply don’t exist in the same way when prompts are portable.
For vendors, the companies that win won’t be the ones with the “best AI.” They’ll be the ones that build the best everything else — infrastructure, integrations, data moats, vertical expertise, and genuine customer relationships that go deeper than a system prompt.
For buyers, you’re entering a golden age of leverage and optionality. Use it wisely, but use it.
The age of the three-year contract and 95%+ gross retention in AI agents may never arrive. And both sides of the table need to plan accordingly.
This wave of churn is coming. The question is whether you’ll be ready for it.
