The AI Became the Commodity. Here’s What 6 Verticals Agreed Was the Actual Moat at SaaStr AI 2026

The six sessions that closed out SaaStr AI 2026:

1. Shoplazza and Subotiz (commerce), Adam Modsley, CRO | The data is the moat, not the AI
2. Nue (revenue operations), James MacArthur, VP of Product Advocacy | AI speed inside deterministic guardrails
3. Papaya Global (global payroll compliance), Sivanne Fishel, VP Client Success, and Hagit Ben-Tzur, Head of Product Design | Build the guardrails before the features
4. Reevo (revenue operations), Ali Ghotbi, CRO | Automate the admin, not the relationship
5. Fisent and Launchpad (regulated fintech), Adrian Murray, Founder and CEO of Fisent, with John Huan of Launchpad (Pegasystems) | Lead with the outcome, not the model
6. The Vertical AI Panel: Scale Venture Partners + GC AI + Inspiren (legal and senior care), Jeremy Kaufmann (Scale), Cecilia Ziniti (GC AI), and Alex Hejnosz (Inspiren) | Where the moat really is

Six companies, six unrelated categories. None of them sold the same product. Almost every one of them ended up at the same conclusion. Here are the top five lessons from each session, in their own context, and then what they all add up to.

Shoplazza and Subotiz: The data is the moat, not the AI

The commerce CRO opened with a platform that builds a working store from a single sentence, then layers seven agents on top for images, ads, payments, and operations across 650,000 merchants and billions in volume.

1. Everyone has the same tools now. Lovable, Claude, Vercel, all of it. Having the tools does not make you successful. The data and making it work together does.
2. Generic context plus a generic prompt gets you generic results. Be specific or get nothing useful back.
3. If you just bolt AI onto your product, you risk becoming a feature inside someone else’s platform. They rebuilt the whole stack AI-first rather than add AI to the old one.
4. A shared, continuous data layer across the entire stack beats stitching point solutions together with MCP endpoints. Magic impresses once. Systems compound.
5. Build for monetization from day one. The cautionary tale: a founder whose AI token deal ran out in month four, got the real bill, and realized they were losing money on every customer. Usage-based billing is not optional anymore.

Nue: AI speed inside deterministic guardrails

Nue is a Salesforce-native CPQ-to-billing platform. The demo built three quote variations in seconds, a task that normally costs a rep two hours.

1. Meet users where they already work. Reps never leave Salesforce, and admins control access through the same permission sets and profiles they always used.
2. The AI is deterministic. Same inputs, same output every time. It asks clarifying questions and never assumes, which is what makes it safe to put in front of a sales team.
3. Don’t generate data nobody uses. Instead of creating three throwaway quotes, it previewed and compared them, and only materialized the one the rep actually wanted.
4. Guardrails live in the pricing engine, not the prompt. Discount floors and ceilings are enforced at the line-item level. Ask for a 76% discount and it caps at 55%.
5. Quote-to-cash has to be end to end. What you quote is what you bill, down to showing the customer their exact first invoice at quote time, with proration accurate to a single day.

Papaya Global: Build the guardrails before the features

Papaya built a compliance AI called Papaya 1 so their clients across 160 countries stop asking ChatGPT a German termination question at 2am and acting on a confident wrong answer that can cost $250,000.

1. The problem is not which model you pick. They gave the same Brazilian employment contract to Claude and ChatGPT. Both were confident, both gave different answers, and neither got it fully right against the actual law. No model had been taught to think about compliance.
2. Turn every failure into a rule. They built 22 rules one at a time, then added a second AI to check the first. An analyst applies the rules like a junior lawyer, a reviewer catches overconfidence and jurisdiction mixing like a senior, and a finalizer ships it like a partner signing off.
3. Build the guardrails before the features. The kill switch is real: if accuracy drops below a threshold in any country, they turn that country off until it is fixed. They pull the plug rather than patch as they go.
4. Trust takes far longer than the build. The agent worked in four weeks. Earning enough trust to put the company’s name on it took four months. They launched to five to ten trusted clients, not everyone, and measured trust by return usage, harder questions over time, and fewer messages forwarded to outside counsel.
5. You can ship production software with no engineers and no UX designers. The build flow was Claude for design exploration, then Claude Code, then Figma’s MCP, then Lovable to deploy, with Supabase for auth, database, and edge functions. The real work was the compliance methodology, not the code.

Reevo: Automate the admin, not the relationship

Reevo aimed agents at the 70 to 80 percent of a seller’s day that goes to administrative work, leaving the relationship work to the human.

1. Start where the time actually goes. Sellers spend most of their day on research, prep, notes, follow-ups, and CRM, not selling. That is the layer to automate first.
2. Meeting prep becomes a living document. It is tied to the calendar, pulls public and in-platform signals, and refreshes itself as new emails and conversations come in.
3. The deal-progression agent does the work, not just the suggestion. It reads the CRM, emails, and call transcripts, cites the evidence (an unanswered “I need to run this by finance” comment), and drafts the personalized recovery email. The rep just hits send.
4. Keep human oversight on the decisions that matter. The CRM hygiene agent fills fields overnight, and the disqualification agent surfaces dead deals with cited evidence, but asks the rep to approve before anything is closed.
5. The productivity numbers are real. Sellers became five times more productive, going from 10 to 15 opportunities each to 50 to 75, with zero leakage. The team hit its number with half the reps while still behind on hiring.

Fisent and Launchpad: Lead with the outcome, not the model

Fisent (content intelligence) and Launchpad from Pegasystems (deterministic workflow) sell what they call outcomes as a service into regulated banks, insurers, and fintechs.

1. Sell the outcome, not the technology. The deliverable is an audit, a claim, or a closed sale, not “AI.” Customers buy the result.
2. Split the work by what should and should not be creative. Deterministic logic runs the workflow and the rules. Generative AI handles ingestion and interpretation. That minimizes variability where variability is dangerous.
3. Procurement does not buy AI because it is interesting. They buy what survives information security reviews, architecture assessments, and compliance audits. Build for the gauntlet, because that is what gates production.
4. Watch your token spend. Don’t burn tokens running the core business workflow. Spend them only where you genuinely need interpretation. Enterprise budgets are getting constrained and buyers are paying attention to this.
5. It is not all-or-nothing agentic. The combination of deterministic workflow and probabilistic model is what actually gets a regulated buyer to adopt and scale. The proof was a dozen customers, 20-plus use cases in production, and hundreds of thousands of pages processed monthly.

The Vertical AI Panel: Where the moat really is

Scale Venture’s Jeremy Kaufmann ran a panel with the founders of GC AI (legal) and Inspiren (senior living) on the state of vertical AI in 2026.

1. AI lets you leapfrog century-old incumbents. GC AI went straight at Lexis and Thomson Reuters, sold to in-house teams instead of law firms, went product-led, and did $100,000 in revenue in its first month.
2. The second mover, or the fifteenth, can still win. Inspiren entered late but viewed the problem through a wider aperture than the early movers, and now has more ARR and faster growth than anyone in the category.
3. Eat the labor, but use the customer’s language. Technology tends to capture 5 to 15 percent of the labor cost it displaces. GC AI delivers a 14 percent cut in outside counsel bills. You sell “do more,” not “fire your team,” and you price against the bigger budget you are displacing.
4. Don’t fight the system of record early. Inspiren feeds the EMR rather than trying to replace it, positioning itself as the system of action. The advice: own at least one of deep workflow, deep data, or outcomes, and play nicely with the incumbent until you have the penetration to reconsider.
5. In GenAI, “better” is still a moat. People pay more for a better lawyer or doctor, and they will pay more for a better product. The moat is customer obsession, a relentless feedback loop, and scale (1,600 companies on GC AI daily). For physical AI, it is deployed hardware plus a falling sensor cost plus a data flywheel. And don’t let the manic swings of AI Twitter rock your boat.

What All Six Sessions Agreed On

Strip away the verticals and the same conclusion shows up in every talk: the model is now the commodity, and the moat lives somewhere else.  At least for these B2B leaders.

Two years ago the pitch was “we have AI.” This year, having AI bought you nothing. Everyone in the room had Claude, OpenAI, Lovable, and the rest. What separated the companies winning real revenue from the ones renting their position was what they built on top of the model.

Four things came up again and again:

• The domain knowledge is the fuel. Papaya said it most plainly: you can copy the engine, you can’t copy the fuel. The 22 compliance rules, the 160 countries of experience, GC AI’s feedback loop across 1,600 customers, Shoplazza’s shared data layer. None of that ships with the model.
• In regulated and enterprise markets, deterministic workflow plus a probabilistic model beats pure agentic, every time. Nue, Fisent, Launchpad, and Papaya’s three-stage pipeline all wrapped the model in workflow the buyer could trust and audit. Lead with the outcome, because procurement does not buy interesting.
• Guardrails come before scale. The kill switch, the discount ceilings, the human approval on close-lost, the launch to ten trusted clients first. The agent is the fast part. Earning the right to put your name on it is the slow part, and it is the part that wins the account.
• Pricing follows the value you displace, not the software category you came from. The admin layer pays off first (Reevo’s 5x), and the bigger budgets, outside counsel, headcount, net operating income, are where the real expansion lives.

The companies that won the room were not the ones with the best model access. Everyone had that. They won on what they did with it.