Every thin workflow SaaS should expect to lose its next renewal to a weekend project

Somebody in marketing at Ramp built their own CRM with more than 50 fields and filters on a single screen. Gergely, who writes the Pragmatic Engineer newsletter and saw it on a visit to Ramp's New York office, found it overwhelming. I had the same reaction, and then a second one: somebody used to get paid for that screen, and now nobody does.

The tool behind it is Glass, which Ramp's four-person team built as what Gergely calls an internal Lovable. Non-engineers plug in internal systems as connectors and generate apps that colleagues can use and modify. Two engineers put the first version together in a matter of weeks. Stripe's equivalent, an AI-assisted prototyping tool called Harbor, was built by one engineer in about six weeks. At Ramp, building a simple tool in Glass is now part of new-hire onboarding, and according to the article pretty much every non-engineer uses it.

Run that through a vendor's renewal spreadsheet. On one side sits a multi-year contract for a workflow product. On the other sits a few weeks of one or two engineers at a company that already employs hundreds of them. The article makes a point I expect procurement teams to repeat back to sales reps for years: picking the right vendor, once you count compliance reviews and price negotiations, costs roughly the same effort as building the thing. For a decade, procurement friction protected incumbents, because switching was painful. Now the same compliance reviews and negotiations make a build look cheap by comparison.

The exposed products are the thin ones. If your product is a table, a form, some filters and a couple of integrations, a Glass-style tool can produce a passable copy wired straight into the customer's own data and styled with its own design library, two things the article says third-party tools handle badly. Team-level CRMs, approval trackers, internal dashboards and prototyping tools are first in line. The irony is that the vibe-coding vendors themselves are on the list. Lovable says it reached about $100M in annual recurring revenue roughly eight months after launch and raised a $200M Series A at a $1.8B valuation in July 2025. Its richest potential enterprise buyers are the companies with enough engineers to build their own Lovable in a month and a half.

Systems of record are safer, and Klarna is the useful case. Sebastian Siemiatkowski said in 2024 that Klarna had shut down Salesforce and Workday. In March 2025 he walked that back on X: Klarna had consolidated its data, partly on Neo4j, rather than swapping SaaS for an AI-built app, and he said he didn't think it was the end of Salesforce. I read that as a boundary marker. The database of record survives. The dozens of small tools bolted around it, each with its own per-seat invoice, are what gets vibe-coded away.

The obvious pushback is that internal builds fail. MIT NANDA's report from August 2025 found that about 95% of enterprise generative-AI pilots produced no measurable profit-and-loss impact, and that internal builds succeeded roughly a third as often as tools bought from specialist vendors, which worked about 67% of the time. Veracode found security flaws in about 45% of AI-generated code samples. And everyone remembers Replit's agent deleting Jason Lemkin's production database during a code freeze in July 2025.

I take those numbers seriously, and I still think vendors are misreading them. NANDA was measuring pilots that were supposed to move a P&L. A marketer's CRM only has to beat the licence it replaces for one team, and it runs on internal data inside the company's perimeter, which is exactly why Ramp and Stripe built rather than handing their data to an outside platform. The Lemkin incident happened on a vendor's product, so it hardly makes the case for buying more of them. More importantly, an internal build hurts the vendor even when it is mediocre. Once every employee at a customer has built something in onboarding, your renewal meeting has a competitor in the room that costs nothing per seat and can be demoed by the buyer's own staff. A buyer with somewhere to walk to negotiates a lower renewal price.

Ramp has now launched Glass as an external product. A fintech company has turned its internal tool into a rival to Lovable and Retool, built by a team of four. If Ramp can sell that, every customer with a coding-agent harness in its cloud can at least threaten to build it.

If you sell workflow software, pull your renewals for the next two quarters and mark every account with hundreds of engineers and every product whose usage sits mostly with one team on one screen. Where both are true, price the renewal as though the customer already holds a quote for one engineer and six weeks, because Stripe has shown that is what it costs.

Prompted by The Pulse: new trend of building “internal vibe-coding apps” at tech companies, newsletter.pragmaticengineer.com.