What Customer Success Means at AI-First Companies — Notes from May 2026

Working notes, May 2026. None of this is settled — I'm writing it down to think more clearly, and because the people I'd most want to talk to about it (founders and deployment teams at AI-first companies) are figuring it out at the same time I am.

---

On May 4, 2026, Anthropic and OpenAI announced enterprise AI deployment ventures on the same day. Anthropic, with Goldman Sachs, Blackstone, and Hellman & Friedman: $1.5B. OpenAI, with TPG, Brookfield, and Bain Capital: $4B at a $10B valuation. Both ventures had the same core idea — embed engineers inside customer organizations and build the AI implementations directly into their operations.

Notice what's not in either announcement. Not "we're scaling our Customer Success team." Not "we're hiring CSMs to manage the deployments." The roles being staffed are Forward Deployed Engineers and Deployment Strategists.

This is the part I want to write about — not because the word "Customer Success" matters, but because the structural change underneath the title change matters a lot, and I haven't seen anyone write about it from the operator's seat yet.

The core thesis I'm working from

The core mission of Customer Success hasn't changed. It still exists to do what it has always existed to do: ensure the customer actually realizes value from the product.

But at AI-first companies, almost everything about how that mission is delivered has changed. The organization, the role, the contract, the measurement, the comp band, the career path — none of it looks like the CS function that B2B SaaS spent the last fifteen years building.

I'm going to be specific about what I mean, because abstract claims about "the future of CS" deserve to be ignored.

What's actually different

1. The org structure looks like Palantir, not Gainsight

Palantir invented the model AI-first companies are now copying: Forward Deployed Engineer + Deployment Strategist, both embedded with the customer, often working on one account at a time. The FDE writes the code that makes the platform work in the customer's environment. The DS owns the operational change, the stakeholders, the adoption. Together they cover what a traditional B2B SaaS company splits across Solutions Engineering, Implementation, CSM, and AM.

The data on what AI-first companies are actually hiring backs this up:

• Anthropic's Applied AI team has FDE openings explicitly described as "embedding directly with strategic customers."
• OpenAI's new DeployCo subsidiary will, in their own words, "use embedded engineers specialized in AI deployment into organizations."
• Dust has a job posting for "Founding AI Deployment Strategist, Post-Sales."
• Mistral has "AI Deployment Strategist, Cybersecurity."
• EY (consulting, not a vendor) launched a UK & Ireland FDE practice in April 2026 — the first major consultancy to formally adopt the model.

FDE job postings rose more than 800% from January to September 2025. Average US compensation: $238k. Staff-level: $630k+.

There is no equivalent CSM hiring boom at these companies. The word "CSM" appears almost nowhere on their job boards. That's not an accident.

2. The contract is outcome-based, not subscription

Sierra — one of the fastest-growing AI agent companies — prices on resolved outcomes. The customer pays only when the AI successfully resolves an interaction, completes an upsell, or saves a cancellation. Sierra went from zero to over $150M ARR by February 2026 on this pricing model.

Adobe announced outcome-based pricing for Adobe CX Enterprise. Decagon charges per resolution. Fewer than 10% of AI companies use outcome-based pricing today, but it's widely expected to become the dominant model for agentic AI products. (Bessemer, Monetizely)

The implication for the post-sales function is enormous. The CSM's traditional job — "drive adoption so the customer renews their seats" — doesn't translate. There are no seats. The revenue scales directly with how often the AI succeeds. The post-sales function exists to grow outcome volume, which is closer to a Revenue Operations job than a relationship management job.

3. The retention numbers are catastrophically worse

This is the part nobody wants to talk about publicly, but the data is in the open.

a16z published their "Retention Is All You Need" analysis of hundreds of AI-native companies. The numbers:

• AI-native overall: 40% gross revenue retention. 48% net revenue retention.
• For comparison: B2B SaaS median is ~82% NRR.

At the high end (above $250/month price points), AI-native NRR climbs to 85% — close to SaaS territory but still below. At low price points (under $50/month), GRR sits at 23%. The analyst Cassie Young at Primary Venture Partners has been calling this the "gross retention apocalypse" — and her argument is structural, not cyclical. Switching costs in AI are lower than in any software category that came before. Prompts are portable. (Growth Unhinged, SaaStr)

I think this is the single most important fact about CS at AI-first companies, and it explains every other structural change.

You cannot run a traditional CSM-led, per-seat-priced CS motion when your category-wide NRR is 48%. The unit economics don't work. You have to either:

• (a) push ACVs up dramatically so the per-account economics support a high-touch motion,
• (b) build genuine switching cost through deep operational integration (which is what FDEs are for), or
• (c) tie revenue directly to outcomes so you only collect when value is delivered.

The AI-first companies are doing all three at once.

4. Measurement looks different

The 2022 CSM dashboard — adoption %, NPS, health score, renewal rate — doesn't survive contact with an outcome-priced agent product. What replaces it:

• Outcome volume and quality — because that's literally what's being billed.
• M3-rebased cohort retention instead of raw retention. a16z's recommended method, to filter out "AI tourists" who try, hit one moment, then churn before the product can compound.
• The "smiling curve" — churned customers returning as the product gets better. This is a real phenomenon a16z documented at multiple AI-native companies. It implies retention is product-trajectory-driven, not relationship-driven.
• Switching-cost depth — how integrated the product is into the customer's operations. This is the moat. The FDE's job is to build it.

The dashboard has become less about the customer's behavior and more about the product's trajectory and the integration depth. That's a quiet but huge shift in what the post-sales function is responsible for.

What I'm seeing from my seat

Some of this lands differently when you're inside it instead of reading about it.

I'm not at a pure AI-first company — the products I've worked on went Seed → C as traditional B2B SaaS — but I am at companies whose product became an AI product over the last few years, and the conversations with customers changed underneath me.

Three things I notice that the industry conversation captures imperfectly:

1. Customers don't know how to evaluate AI products. Sales sells outcomes. The customer signs. In month one, they hit you with: "How will we actually measure this?" The honest answer is: we have to design the eval together. This used to be a Product question. Now it's a post-sales question. None of my CS playbooks from 2019 have a chapter for it.

2. Quality regressions are real and they're CS tickets. A model updates. A prompt changes. The customer's success rate drops 4 percentage points. They notice. They open a ticket. Whose ticket is it? Engineering's? Product's? CS's? Every team I've talked to has improvised an answer. I don't know that anyone has the right one.

3. Usage going down is sometimes the win. When the AI gets better, agents need fewer turns. Customers ask fewer follow-up questions. Tickets resolve in shorter sessions. The 2022 health score (which weights usage positively) reads this as a churn signal. The 2026 reality is the opposite. I've had to argue this point with my own dashboards, which is a strange experience.

What this means for people doing CS today

If you're in CS at a traditional B2B SaaS company, here's the honest read:

The function you have been doing isn't going away — there are still thousands of companies that need exactly the role as it was. But the highest-leverage version of CS in 2026 is migrating toward the FDE / Deployment Strategist model at AI-first companies. The career trajectory, the comp, the equity, the proximity to interesting problems — all of it is concentrating there.

The people who can credibly take those roles are not pure-CSM profiles. They are people who can hold two halves:

• Commercial fluency + outcome ownership (the traditional CS strength)
• Technical depth that lets them participate in the engineering side (the new requirement)

That's not most CS profiles. It's a small population. If you can credibly do both, the demand for you in the next 18 months will be unusual.

What I'm still figuring out

I'm trying to be honest about what I don't know. Some open questions:

1. Is the FDE/DS model going to stay this dominant, or does "Customer Success" come back as a separate function once AI-first companies mature past the early-deployment phase?
2. What's the right post-sales staffing model below roughly $1M ACV? Palantir-style 1:1 embedding only works at very high contract sizes. The math for mid-market AI is wide open.
3. Where does eval co-design with the customer actually live, organizationally? Today it's CS-adjacent at most AI-first companies but nobody owns it cleanly.
4. How do outcome-based contracts get renewed when the AI's capability shifts every quarter? Sierra, Decagon, and Adobe are working this out in production.
5. Is there a clean career bridge for current CSMs into Deployment Strategist roles, or is the technical bar high enough that the population is locked out?

What I'd love pushback on

If you work in post-sales at an AI-first company — Anthropic, OpenAI, Dust, Sierra, Decagon, Glean, Cresta, Mistral, or any of the others — and any of this reads wrong, I want to hear it. I'm writing from the seat of someone who has done CS at scale for a product that became an AI product, not from inside an AI-first org. The view from the inside is the one I'd most like to correct against.

Specifically:

• Am I overstating the death of the CSM title? Are there AI-first companies where the function is alive and well under that name?
• The retention numbers — do they match what you're seeing day-to-day, or is the public data lagging the operational reality?

I'll keep these notes updated as I learn.

---

This is the first in an ongoing series of working notes on Customer Success in the AI era. The next pieces will go deeper on specific stages of the post-sales motion: handoff, kickoff, eval design, QBR replacement, and renewal mechanics.

If you want to be notified when those go up, subscribe to the newsletter — and if you want to argue with anything in here, my LinkedIn DMs are open.