If You Are at Scale, You Now Have 2 Full-Time Jobs: Keep Your Installed Base Happy. And Winning the AI Agent War in Your Space. | Saa Str

Overview

AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

AI Agent Playbook Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

Details

University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

AI Annual 2026 Events Overview Sponsors

	Event Sponsorship
	Media Sponsorship

Digital AI Day 2026 (Free) Speaker Submissions Speaker Requirements Overview

If You Are at Scale, You Now Have 2 Full-Time Jobs: Keep Your Installed Base Happy. And Winning the AI Agent War in Your Space.

by Jason Lemkin | Artificial Intelligence (AI), Blog Posts, Saa Str. Ai

Here’s what I see at almost every B2B company that has crossed $50M ARR right now:

The CEO is running two completely different, completely demanding businesses at the same time. And most of them are only really showing up for one.

The first job is the one they know. Keep the installed base happy. Keep NRR above 110%. Keep churn below 5%. Grow revenue from the customers you already have. It sounds simple. In practice, at any real scale, it will consume every waking hour if you let it.

The second job is new. Build and deploy the #1 AI agent in your category. Not a decent one. Not a “we have AI features too” one. The one that makes prospects say: “Oh, we’re using them because they’ve got the best AI.”

These are not the same job. They do not overlap much. And you cannot afford to lose at either one.

Keeping your installed base happy at scale is a full-time job before AI, before platform risk, before anything.

Think about what it actually requires at

You have hundreds or thousands of customers across different segments, all at different stages of adoption, all with different success metrics. Your enterprise customers from 2019 want features your 2023 mid-market customers have never heard of. Your fastest-growing customers are pushing your product in directions your roadmap didn’t anticipate. Your oldest customers have workarounds and integrations built on top of behavior you are about to change.

Every quarter, someone on your board asks why NRR is “only” 112% when it was 118% two years ago. Your CS team is understaffed relative to the account load. Your support tickets keep growing faster than your revenue. Your biggest customers want executive relationships you can barely sustain.

And now, on top of all of that, your customers are getting pitched weekly by AI-native competitors who promise to do 80%-200% of what you do (sometimes at 40% of the price), with a better interface, and no legacy complexity.

CIOs are looking to reduce existing vendors to make room for new AI vendors.  That means everyone is under extra scrutiny at renewal time.  CIOs are trying to hold net vendor counts flat.

The pressure on the installed base has never been higher. Customers are scrutinizing every renewal. They are benchmarking you against tools they built internally in four weeks on top of an LLM API. They are asking harder questions about ROI than they did in 2021 when everyone was throwing software at every problem.

Just holding NRR at scale right now is hard work. It can consume the entire team, including sales and GTM. The companies doing it well are investing heavily in customer success, in onboarding, in making their product so deeply embedded in workflows that switching is genuinely painful. They are doing QBRs that actually review outcomes, not just usage stats. They are building CS motions that separate expansion from relationship, so customers do not feel every conversation is a sales pitch dressed up as a check-in.

We have seen what happens when companies neglect this and treat the installed base as a guaranteed revenue stream to extract from. They jack up prices without delivering more value. They cut CS headcount to hit EBITDA targets. They turn “customer success” into a collections department for expansion revenue.

It works for a quarter or two. Then NRR starts sliding. Then churn accelerates. Then the brand takes a hit because unhappy enterprise customers talk to each other. Then new logo acquisition gets harder because references dry up.

At scale, your installed base is your most valuable asset and your biggest ongoing responsibility. It does not manage itself.

And now it has all gotten much harder.  Harder than ever.

While you are focused on keeping 500 enterprise customers happy, several well-funded teams of 15 people is building AI-native versions of your product.

They have no legacy code. No enterprise support overhead. No installed base to protect. They are moving fast, they are building on top of the best models, and they are deploying agents that can do workflows your product still requires humans to manage.

If they build the best AI agent in your space before you do, you have a serious problem.

Not a “we should think about our AI roadmap” problem. A real, existential problem.

Buyers in most B2B categories now have an AI requirement. It is not a nice-to-have. Enterprise procurement teams are asking about it in RFPs. Mid-market buyers are comparing AI capabilities as a primary filter, not a secondary one. SMBs are increasingly choosing tools based on what they can automate, not just what features they get.

If you are not the clear #1 AI agent in your space, you are losing deals you used to win automatically. And you are losing renewals to companies that can show a customer they replaced three human workflows with one agent.

Customers do prefers to buy AI Agents from their existing providers, the latest Redpoint CIO survey shows that.  But if it’s not there, they will buy from new emergent leaders.

The bar is not “we have AI features.” The bar is: when someone in your target market says “which tool has the best AI for [your category],” your name comes up first. That is the only position worth holding.

What 141 CIOs and $765 Billion in Capex Tell Us About Where B2B Software Is Headed: The Latest From Redpoint

What 141 CIOs and $765 Billion in Capex Tell Us About Where B2B Software Is Headed: The Latest From Redpoint

The Resistance Is Coming From Everywhere. Including Your Own Team.

Your installed base will actively resist your AI buildout. Not maliciously. They just want what they bought.

Think about the reality of your average enterprise customer in 2026. They signed a three-year contract. They spent six months implementing your product. They trained their team on it. They built internal workflows around it. They have quarterly business reviews, integration dependencies, and approval chains tied to how your product works today.

When you start shipping AI features that rethink core workflows, you will hear this:

“We don’t need that AI agent right now.  It would be nice, but right now we just need the reporting dashboard to load faster.”

“Can you focus on fixing the thing that’s been broken since Q3 before adding new stuff?”

“Our compliance team needs to approve any AI features before we can use them, so please just keep shipping the product we bought.”

Every single one of those requests is completely reasonable. And responding to them is Job #1. It is not optional. If you ignore the 80% of your customers who just want what they bought to work well and get better incrementally, you will destroy your NRR and your reputation.

But here is the trap: if you only listen to those customers, you will lose the category. Because the 20% who want AI capabilities are your future. And the AI-native competitors coming for your space are building specifically for them.

The installed base will consume your roadmap, your CS team, your support org, and your engineering cycles if you let it. Not because your customers are wrong. But because at scale, the legitimate operational needs of thousands of customers will always outweigh the speculative investment in a new product direction — unless you explicitly protect that investment.

Your VP of Product will tell you the AI agent is cannibalizing existing customers. They are not wrong about the risk. They are wrong about what happens if you don’t do it.

Your head of Engineering will tell you they cannot pull 10-15 senior engineers off the core product. Those engineers are handling incidents, shipping features for your largest customers, managing technical debt. Every one of those things is real and urgent.

Your Sales leaders will worry about what a transformational AI product announcement does to in-flight deals. “We just told a prospect our current product is the answer. Now we’re saying AI is the future?” It creates confusion at the worst possible time.

Your CS team will flag that the workflows customers depend on today might break. They are right.

These are not obstructionist people. They are good operators doing their jobs. They are the antibodies your company developed to protect the product you have. And those same antibodies will slow-walk your AI buildout to death through a thousand small objections if you let the resistance go unmanaged.

The companies that are winning both jobs are not winning because they ignored the resistance. They are winning because they anticipated it, named it directly at the leadership level, and made explicit decisions about where the organization had permission to push through it.

That means the CEO has to be the one who holds both jobs simultaneously and refuses to let one consume the other. It cannot be delegated. The head of Engineering cannot make the tradeoff between core product stability and AI agent velocity. Only the CEO can hold that tension and decide where the resource allocation lands.

This is the trap I see so many companies have fallen into

They focus so hard on protecting NRR that they under-invest in AI product velocity. They rationalize it: “Our customers are happy, our retention is strong, we will get to AI next quarter.” Meanwhile the category is shifting under them.

Or they swing the other way. They announce a big “AI-first transformation,” redirect engineering resources, ship a flashy new AI product — and then their existing customers feel abandoned. Implementation quality drops. Support degrades. The customers who made you successful start looking around.

You cannot sacrifice one to fund the other. Both have to work.

The companies navigating this well are doing a few specific things:

They are treating the installed base as a distribution advantage, not a defensive burden. Every existing customer is a potential power user for your AI agent. They already trust you. They already have their data in your system. If you build the right AI capabilities on top of your core product, your installed base is the fastest path to AI adoption at scale. This is a massive advantage over AI-native startups with zero customers. Use it.

They are using customer feedback from the installed base to win the AI product war. Your best enterprise customers will tell you exactly what workflows they want automated if you ask the right questions. The companies building the best AI agents in their categories are not guessing. They are building on top of deep knowledge of how real customers work. Your installed base is a four-year competitive intelligence operation if you treat it right.

They are honest internally about what “winning at AI” actually means. It does not mean a chatbot on your website. It does not mean adding GPT-4 to your search bar. It means your product can take over a meaningful workflow end-to-end, with results good enough that a customer would rather use your agent than hire a person to do the same task. That is the bar. Every company should be asking themselves honestly whether they are building to that bar or building to a demo.

They are moving fast on AI without breaking the core product contract. Your installed base has expectations. They rely on certain behaviors. Reliability, predictability, data integrity. Your AI buildout cannot break those. The best operators are running parallel tracks: AI product velocity on new capabilities, ironclad stability on existing ones.

The Trap: If You Are Not Careful, Your Installed Base Becomes the Thing That Kills You

We talk about the installed base as an asset. It is. But we spent a lot of time on the 20VC x Saa Str pod the other day on the flip side: how your installed base, if you let it, becomes the thing that prevents you from winning in AI agents at all.

"In many ways, having

"In many ways, having

Because supporting it, servicing it, patching it, fixing bugs, closing feature gaps … can seem to suck up 100% of the team's time.

And if you aren't very… https://t.co/USOr 6 Glf To pic.twitter.com/p 4 Rfeg 6NPE

— Jason ✨👾Saa Str. Ai✨ Lemkin (@jasonlk) March 26, 2026

You have 1,000 customers. They pay real money. They have real operational dependencies on your product. Their requests are constant, specific, and legitimate. So you respond to them. You staff up CS to handle their needs. You let their roadmap feedback shape your engineering priorities. You take the 6-8% annual price increases that the base will bear and call it growth. Your board meetings are mostly about NRR and churn and expansion revenue. You feel like you are running a tight, well-managed business.

Because while you were doing all of that, three people who have never had a single customer walked out of Stanford and built something on Claude that does the core workflow your 1,000 customers pay you for. And it is better. And it is cheaper. And it is faster to deploy. And it requires no implementation.

This is the conversation that came up repeatedly on 20VC. Rory put it exactly right: there is a race between incumbents who have distribution and need to add AI product, and new entrants who have AI product and need to add distribution. How long your distribution advantage actually lasts depends entirely on how fast the product gap opens up — and how fast new entrants can close the distribution gap.

The answer, in most categories right now, is: faster than you think.

Look at Figma.  It Won Its

$1B 2014- 2024 Category.  But It Lost All The$

500m+ Of Product Revenue Repit and Lovable Have Already Closed.

By almost any measure, Figma is an exceptional company. Best-run design software in the world. Passionate installed base. Strong growth, strong margins. And yet Replit and Lovable — companies that barely existed two years ago — built the adjacent AI prototyping and building category that Figma had every right to own. The honest conversation inside Figma should be: “We owned this space. We let it go.” The installed base didn’t cause that. But the weight of serving the installed base, and the organizational gravitational pull toward improving the existing product over betting on an adjacent one, contributed to it.

Or Intercom. Eoghan came back as CEO and made a decision that most professional managers at established companies cannot bring themselves to make: he gutted the existing product architecture and bet the company on Finn. For a year, his best customers were asking why he was not focused on the real business. His CS team was managing confused enterprise accounts. His sales team had to explain a product that was effectively being rebuilt in public.

That is what it actually costs to not let the installed base become a trap. You have to be willing to make your existing customers uncomfortable. To tell them, directly, that the future of your product is not the same as the past. To ship things that disrupt the workflows they have built on top of your old architecture.

Most companies cannot do it. Not because they are run by bad people. Because the installed base is a living veto on bold product decisions. Every enterprise customer who says “please don’t change the API” is a reasonable person with a reasonable request. But if you honor every one of those requests, you end up with a product that is optimized for how your customers worked two years ago.

The other trap is subtler and more dangerous: your installed base teaches you the wrong lessons about what to build in AI.

Your best customers will tell you what they want. They want the features they almost have. They want their current workflows automated. They want incremental improvements on the product they bought. They are not going to tell you to build something that makes them reconsider whether they need your product at all.

So if you survey your installed base and build your AI roadmap from those answers, you will build a perfectly adequate AI-augmented version of what you already have. And you will lose to someone building a truly AI-native version of what your customers actually need to do.

This is the trap. The installed base is both your greatest asset and your greatest source of inertia. It funds the company. It gives you distribution. It gives you data. But it also tells you what it wants, and what it wants is almost never the thing that wins.

That takes a kind of organizational courage that is very hard to sustain when your board meeting is also a conversation about why NRR dropped two points.

Which is exactly why this is two full-time jobs. Not one job with an AI feature attached.

Here’s What It Looks Like From the Other Side of the Table

I can tell you what this feels like as a customer. Because Saa Str is both a company trying to win Job #2 and a buyer evaluating every established vendor’s attempt to do the same.

We run 20+ AI agents at Saa Str. They power our outbound, inbound, content, customer support, operations, and event management. We operate at eight-figure revenue with a handful of humans. The AI stack is not a pilot or an experiment. It is how we run the company.

Here is the part that should concern every established B2B company reading this: exactly one of our 20+ AI Agents agents comes from a legacy vendor.

That vendor is Salesforce, with Agentforce. And to their credit, it works. Because Agentforce has something the AI-native point solutions don’t: it knows everything your Salesforce already knows. Every past interaction. Every company record. Every note, every event attendance, every prior deal. That native data access produces results. We’re seeing 72% open rates on leads that had been completely ghosted, from contacts that got zero human follow-up for months. It is the highest response rate of any AI platform we run.

But it took Salesforce years and billions of dollars to build that. And it is still just one agent.

Everything else we run — the AI SDR sending 15,000+ messages with 5-7% response rates, the AI BDR pre-booking inbound meetings with full Salesforce and Marketo sync, the AI for call transcription and CRM auto-fill, the AI for content, the AI for customer support — all of it comes from companies that either did not exist or were tiny startups when most of our legacy vendors were at their peak.

This is not a fluke. It is a pattern we hear from execs everywhere.

Ask a VP of Sales at a $200M ARR company which AI SDR they deployed. It is not from their CRM vendor. Ask a CMO where their best AI content infrastructure came from. It is not from their marketing automation platform. Ask a Head of Customer Success which AI tools are actually automating ticket resolution. Almost never from their primary CS platform.

The established players had the advantage. They had the customer relationships, the data, the distribution, the trust. They should be winning every category by default. Buyers wanted them to win — it would have been so much easier. One vendor. Native integration. No security review for a new tool. No new contract.

But most of them were too slow. Or they bolted AI onto existing products that were not designed for it. Or the political resistance inside their own organizations — the same antibodies I described above — kept them from moving at the speed the market required.

And now their customers are making purchasing decisions that would have been unthinkable three years ago. Deploying half a dozen AI-native tools from companies with 40 employees. Running critical workflows on infrastructure that did not exist 18 months ago.

The window is not closed for established vendors. Salesforce proved that. But it is closing. Every quarter that a category’s AI-native leader gets more customers, more training data, deeper integrations, and more switching costs built up is a quarter harder to take it back.

If you are running an established B2B company, the question is not theoretical. Right now, your own customers are evaluating whether to solve their AI workflows with you — or with someone else. Most of them would prefer it to be you. They trust you. They have your data. They want fewer vendors, not more.

If you are at

But if a well-executed AI-native competitor enters your category and starts winning 20% of your new logo pipeline, and then starts showing up in your renewal conversations, your growth rate compresses fast. Within 18-24 months, you are looking at a fundamentally different business trajectory.

The companies that protect that trajectory are the ones that never let either job slip. They invest in the CS motion that keeps NRR strong, and they invest in the AI product motion that makes the answer to “who has the best AI in this space” obviously, clearly theirs.

That is two full-time jobs. Neither one pauses while you focus on the other.

If you are at scale right now, you already know this in your gut. The companies that act on it are the ones that will look back at 2026 as the year they got it right.

Yes, This Is the Hardest the Job Has Been in 15 Years. Many Are Not Up For It.

Everything described in this post — keeping the installed base genuinely happy while simultaneously building the best AI agent in your category, managing the internal resistance, not letting your customers trap you in your own past — this is harder than running a B2B company has been at any point in the last 15 years. Harder than the 2022 downturn. Harder than the post-2021 reset. Harder than the brutal GFC and March 2020.  Much, much, much harder than navigating the transition from on-premise to cloud.  We had years and years to do that.

CEO resignations are up significantly across the industry right now. Not just forced exits. Voluntary ones. Founders who built real companies, hit real scale, and are stepping back. Operators who have run the business for five or six years and are handing over the keys.

Some of that is normal leadership transition. But a lot of it is something else. A lot of it is founders and CEOs who looked at the two-job reality and made an honest assessment: I am not sure I want to do this. I built a great company. I do not know if I am the right person to rebuild it again from the inside, under fire, while also protecting everything I already built.

That is an honest answer. There is nothing wrong with it.

But if you are going to stay, if you are going to lead the company through this, then the answer has to be different. The answer has to be: I am doing both jobs. Not one of them. Not one and a half of them. Both.

Because the companies that only do Job #1 — protect the installed base, run the existing business well, optimize for NRR — are heading toward a slow decline. The AI-native competitors will take the new logos first. Then they will start showing up in renewal conversations. Then the enterprise contracts get harder to close. Then the brand loses its authority in the category. Then it is very hard to reverse.

And the companies that only do Job #2 — bet everything on AI, neglect the customers who are paying the bills — blow themselves up before the AI product is even ready.

There is no path that avoids doing both. Not at scale. Not right now.

This is the job. It is hard. The leaders who accept that and get to work — who hold both jobs simultaneously, who manage the resistance without being paralyzed by it, who move fast on AI without abandoning the customers who made them — those are the ones who come out the other side with real companies.

The ones who wait, or who let one job crowd out the other, are on a timeline they probably do not realize is running.

Grow or die. It has always been the rule in B2B. Right now, it just has less margin for error than it ever has before.

Palantir: The Greatest Enterprise Software Company Of All Time Ever? Could Be.

Saa Str's AI Agent Playbook: How We Deployed 20+ Agents to Scale 8-Figure Revenue with Single-Digit Headcount

Linked In: The Greatest B2B Acquisition of All Time?

Palantir: The Greatest Enterprise Software Company Of All Time Ever? Could Be.

Saa Str's AI Agent Playbook: How We Deployed 20+ Agents to Scale 8-Figure Revenue with Single-Digit Headcount

Linked In: The Greatest B2B Acquisition of All Time?

Industry News

Get from

Key Takeaways

• AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

• AI Agent Playbook Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

• Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• AI Annual 2026 Events Overview Sponsors

    Event Sponsorship
    Media Sponsorship