I burned an hour scrolling the Show HN thread for “ROI-first AI automation framework for B2B companies” that climbed to #4 this week, and the whole comment section read like eavesdropping inside a server room. Stripe webhooks, Salesforce APIs, n8n versus LangGraph for ETL, Snowflake pipelines, MCP servers, ROI math priced in margin points. Not one mention of the browser. Which is weird, because if you walk into any mid-sized B2B company and watch what people do for eight hours straight, what you mostly see is humans inside Chrome tabs.

The whole “automation framework” category bakes an assumption into its name: every business process is at heart an integration problem, and the only reason humans still touch it manually is that nobody has wired enough services together yet. Not wrong. Just wildly incomplete.

What the framework can’t reach

Pick a real B2B workflow. A sales rep updating a Salesforce opportunity after a call. An AP clerk hunting an invoice PDF inside a vendor portal because it landed in a shared inbox. A procurement person clicking approve inside Coupa on a request with eleven approvers strung across three departments. Someone in HR trying to coax benefits data out of a payroll system that has not shipped a real API since 2014 and almost certainly never will. All “automatable” in some abstract platonic sense. None reachable by LangChain, n8n, or whatever ROI-first thing trends next week.

Most B2B SaaS does not expose what you’d actually want to automate. Vendor portals especially. ADP, Concur, NetSuite-on-an-old-tier, half the healthcare stack, every state procurement system I’ve ever sworn at, every SOC 2 reporting tool I’ve ever logged into. The framework treats as “integrable” the 20% of the surface with a clean REST endpoint, and the rest is screens, clicks, OAuth redirects, and ASP.NET forms with hidden viewstate that fall apart if you tab through fields in the wrong order.

ROI is screen-shaped

The HN post defines ROI in terms of replaced backend tasks. Faster lead enrichment, cheaper invoice reconciliation, lower API egress. Real numbers. But the largest ROI lever in most B2B orgs has never been the integration layer. It’s the interface layer, where some unfortunate human is copy-pasting a number from one tab to another forty times a day because neither vendor cares enough about the use case to ship a proper integration.

So if the framework can’t touch a browser, it can’t touch that work. And that work is the work.

Already logged in versus spin up a session

Browser automation has a quiet split between two architectures. One spins up a headless Chrome on a cloud VM, logs in somehow, runs a script, tears it down. The other runs inside the tab the user is already in, with cookies, SSO, 2FA, and conditional-access state already warm. Dassi lives in the second camp, which is why I keep yelling from this corner.

The cloud approach hits a wall the moment a vendor portal demands SSO through Okta with conditional access, or a SaaS tool sends a magic link to a personal inbox, or a procurement system fingerprints a non-resident browser and pops a security question that your cloud assistant has no business knowing the answer to. We wrote earlier about why cloud browser agents can’t see your tabs, and the same constraint quietly kills “ROI-first” frameworks the second they try to leave the API surface and crash into a real B2B auth wall.

The README test

I read through the GitHub READMEs of roughly a dozen B2B-flavored automation frameworks last weekend, partly out of curiosity and partly because the HN thread had been bugging me all week. Almost every one had an integrations folder: Salesforce, HubSpot, Slack, Linear, Notion, Postgres, Stripe, Zendesk. Barely any had a browser folder. The closest was usually a thin Playwright wrapper buried under “experimental” or “labs,” with three open issues and no maintainer who had pushed in six months. The product surface ends exactly where the documented API ends.

Which is a damn strange line to draw if ROI is supposedly the goal, because the highest-friction work inside B2B is precisely the work without an SDK. We made a similar argument about browser agents being the new API, and the framing keeps holding up. Browser agents are, in practice, the de facto API for every system whose vendor never bothered to ship one.

Tell

When the marquee discussion of “B2B automation” excludes the slice of B2B work that everyone in the org actually does, the framework is solving the wrong problem with admirable efficiency.

The microwave problem

Calling something a “B2B automation framework” when it can only reach the slice of the business that exposes itself programmatically is roughly like calling a microwave a kitchen. Microwaves are useful. They are also one appliance, and most of the cooking happens elsewhere.

Dassi lives in the Chrome side panel for exactly this reason. Backend pipelines are a solved space with five frameworks brawling over a shrinking pie. The browser layer, where a knowledge worker’s afternoon actually goes to die, has roughly nobody serving it well yet.

So if your team is reading the Show HN thread and wondering whether to adopt the trending framework, ask first whether the things eating your team’s hours are API-shaped or screen-shaped. That answer picks the tool. The headline shouldn’t.