Five Arguments for AI-Native CRM and ERP, and What Each Depends On
Founders building AI-native CRM and ERP products make five arguments for replacing Salesforce, NetSuite, and SAP. What each argument depends on, the one I expect to hold, and how a buyer can test a pitch.
Part of The Market
The founders building AI-native CRM and ERP products say plainly that they plan to replace Salesforce, NetSuite, and SAP. Others plan to leave those systems in place and take over their users. “AI-native” means the product was designed around AI agents from the start, with no older product underneath. Investors have backed them with hundreds of millions of dollars. Their confidence rests on five arguments. A buyer who understands those arguments can judge any of these products on its merits.
Who is making the claim
The founders know the incumbents well. Bret Taylor was co-chief executive of Salesforce before he started Sierra, which builds customer service agents. Christopher O’Donnell led the team that built HubSpot’s CRM, and he now runs Day AI. Nicolas Kopp ran the U.S. business of the digital bank N26 before he founded Rillet, an AI-native ERP.
When these founders say an older design cannot do something, they describe systems they built, ran, or depended on. That history is one reason investors take the claim seriously.
Argument one: platform changes create new leaders
Large changes in computing have produced new software leaders before. Salesforce rose when software moved to the cloud, and the on-premise vendors of that period lost ground. The startups see AI as the next change of that size.
Their investors say so directly. Michael McBride of GV, which led Attio’s latest round, notes that CRM has gone 25 years without a major platform change. Bain Capital Ventures wrote that the timing for Rillet is new. “Five years ago, building a new ERP would have been a fool’s errand.”
AI also lowered the cost of building. Coding tools let a small team write far more software than before. A new CRM or ERP is now within reach of a startup.
Argument two: the old products were built for people at screens
The founders make this argument most often. An older CRM or ERP assumes a person will type data into forms and click through menus. An AI agent needs something different. It needs complete data, current data, and a structure it can read without a person in the middle.
Kopp describes Rillet as “built for agents versus humans sort of clicking buttons and navigating around.” He points out that older ERPs calculate financial results on a schedule, in batches. Rillet’s ledger updates with each transaction, so an agent always works from current numbers.
O’Donnell makes the same case for CRM. A traditional CRM depends on salespeople to enter what happened, so the record is only as good as their typing. Day AI captures conversations and builds the record itself. “This is the Cursor for CRM, it’s the same pattern,” he says.
The startups grant that incumbents can add AI features. Their claim is that features on top cannot fix a foundation designed for manual entry. GV made the point when it invested in Attio, writing that “a new underlying architecture is absolutely necessary.”
Argument three: incumbents earn their money from seats
Most business software is sold by the seat, with one license for each person who uses it. Agents do work without a seat. A vendor paid by the seat can earn less as its own agents get better.
Startups have no seat revenue to protect, so many charge for results. Sierra gets paid when its agent resolves a customer’s issue. Taylor argues that incumbents will struggle to follow, because a company built on one way of charging rarely changes it. “Changing your business model is really hard,” he told a Sequoia audience.
The data gives that view some support. AlixPartners analyzed 65 SaaS companies and AI-native competitors and found that only four had fully adopted pricing by results. More than half of the others still relied mainly on seats.
Argument four: the larger prize is the labor budget
Software vendors compete for the software budget. The startups aim at a larger number. Sequoia partner Julien Bek wrote in March that companies spend about six dollars on services for every dollar on software. A product that does the work can compete for the six.
Early examples support the point. Rillet says its customer Mercor runs a business with more than $2 billion in annual recurring revenue on a three-person finance team. Salesforce agreed in June to buy Fin, the company once called Intercom. Salesforce cites cases where Fin’s agent resolves 76% of support volume on average. Both figures come from the companies involved, so treat them as claims until an outside party measures them.
Andreessen Horowitz frames the point another way. An incumbent’s agents work outward from the record it already owns, partner Seema Amble wrote in September. A startup can manage a whole job that crosses several systems. Elon Musk goes further with Macrohard, a joint xAI and Tesla project meant to do a whole software company’s work with agents.
Argument five: start small, then move up
No startup expects a large bank to replace SAP next year. The plan is to win companies that have little to move, then grow with them. Attio’s 5,000 paying customers include young AI companies such as Lovable and Granola. Rillet’s first customers were technology companies, and about 40% of its 600 customers now come from other industries.
The startups expect today’s fast-growing young companies to become tomorrow’s large ones, and to keep the software they started with. Some are already moving up. Rillet says it now sells to more complex customers, including public companies. Half of Sierra’s customers have revenue above $1 billion.
The results so far
The growth figures explain the funding. Menlo Ventures found that startups took 63% of enterprise spending on AI applications in 2025, up from 36% a year earlier. Sierra reached $100 million in annual recurring revenue in seven quarters. Rillet’s new annual recurring revenue doubled in the three months before its August round, when it raised $100 million at a $1 billion valuation.
Investors have turned these results into a general thesis. Bessemer Venture Partners expects AI to reshape the record in every software category within five to ten years. Battery Ventures wrote in September that this is the best time in many years to build a new one.
What each argument depends on
Each argument holds under a condition, and the incumbents are working on every one. The platform argument depends on incumbents being unable to get the new technology. This time they license the same models the startups use. ICONIQ, which surveyed about 300 software executives, summed up the result. “AI is no longer a differentiator on its own.”
The design argument depends on older records staying closed to agents. This fall, Salesforce, Workday, ServiceNow, and NetSuite each connected their records to Claude or ChatGPT, with existing permissions intact. An agent can now work on an older record. The startups’ answer is that the agent still reads data entered by hand and processed in batches.
The pricing argument depends on incumbents refusing to change, and some are changing. ServiceNow said in July that about half of its net new business is not priced by seat.
The labor argument depends on cost. Every agent request spends money on a model, and ICONIQ projects average gross margins of 53% on AI products this year. Traditional software has held a median near 80%. A startup that charges for results must cover the model’s cost on every result.
The move-up argument depends on how hard it is to change systems at a large company. AI is making it easier to move data. Rebuilding approval rules, satisfying auditors, and retraining staff still take time.
Investors are debating these same points. Amble’s piece is titled “The Incumbents Are Coming.” It notes that a customer can now connect a record’s data to Claude and skip the startup’s app. Her advice to startups is to win through focus on one job.
My read
I expect the design condition to break first, and it is already breaking. Four large incumbents opened their records to outside agents this fall, with permissions intact. A buyer can now get agent work on a system it already owns. “Built for agents” is a weaker reason to move than it was a year ago.
Pricing is next. ServiceNow shows that an incumbent can sell by the seat and by usage at once.
The labor argument is the one I expect to hold, and I have seen why from the inside. At Invoiced and then Flywire, I worked on billing, receivables, and payments software for finance teams. The hard part was never the ledger. It was the work around it: chasing invoices, matching payments to them, checking the businesses on both sides, and connecting to each customer’s ERP. Finance teams paid people to do that work, and no system of record did it for them. That is the six dollars.
So startups and incumbents can both win, at different layers. A startup that does a whole job across systems, without asking the buyer to move its record, earns the labor budget. The company that holds the record keeps the account, because moving a ledger at a large company is a project nobody wants to own. A new CRM or ERP will keep winning young companies and growing with them. I do not expect one to replace Salesforce or SAP at a large company on design alone. The large prize goes to whoever does the work on top of the record the customer already trusts.
How to use this
You do not need to share my read to use the five arguments. Ask a startup which of the five arguments its product rests on, then test the condition. If the pitch is design, ask what the product does that an agent connected to your current system cannot. If the pitch is price, compare the cost of each result with what you pay now. If the pitch is labor, ask for a customer you can call.
Ask the same of your current vendor. Find out whether its record is open to outside agents, how it prices agent work, and what it plans to change. The tools you already pay for belong in this comparison, and many now accept agents under the permissions you already set.
Either choice depends on your own preparation. Agents need clean data, written rules, and a named owner for each process. A company that does that work will get results from a new product or an older one.
First published on LinkedIn.
Published .
Cite this
Couch, C. (2026). Five Arguments for AI-Native CRM and ERP, and What Each Depends On. Ends at Cash. https://chriscouch.ai/essays/five-arguments-ai-native-crm-erp/
BibTeX
@misc{couch2026fiveargumentsforainative,
author = {Couch, Chris},
title = {Five Arguments for AI-Native CRM and ERP, and What Each Depends On},
year = {2026},
howpublished = {Ends at Cash},
url = {https://chriscouch.ai/essays/five-arguments-ai-native-crm-erp/}
}