← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Sourceupdated Sat Oct 24 2026 08:00:00 GMT+0800 (Philippine Standard Time)

The End of the App Era What Comes Next (Nate B Jones)

saasagentsenterprise-softwarebuyingvendor-managementdata-layerworkflowsmcpagent-lock-insalesforcemicrosoftmetadoordashworkdaycrustdataciocoo

The End of the App Era: What Comes Next? (Nate B Jones)

A ~30-minute AI News & Strategy Daily essay (clipped 2026-10-24; cases dated to mid/late September 2026) on how agents change what software is for and therefore how it gets bought and sold. Nate B. Jones's answer to the SaaS-is-dead story: not death, decomposition — and the sticky layer has moved from the app to the agent you trained.

Key claims

1. The agent you train gets harder to leave; the tool it replaces gets easier

A recruiter's agent pulls candidate data, drafts shortlists, learns which hiring managers want what and who actually got hired. She stops opening the recruiting tool. Months later a CFO wants to switch AI vendors for cost or optionality, and nobody at the table sees that "weeks and weeks of progress" live in her harness and memory. → Agent Lock-In.

2. What are you still there for?

"When you open an application, what are you still there for? Information you can't get elsewhere. Something you need carried out correctly." SaaS sold a muddy bundle — finding, arranging, acting on and storing information — so an agent getting good at part of it exposes the whole relationship. Incumbents are "reasoning from the data and distribution they already have into the agent position they think they can hold."

3. Three layers: data, application, agentic

Everyone — employee, buyer, seller — now has to state their position against the data layer, the application layer (decomposed into interfaces and workflows) and the agentic layer. Interfaces thin out; compliance-bearing workflows and unique data get more valuable; the agentic layer absorbs the work and becomes the new moat. "I do not think that we will all be using the command line all the time." → Data, Application and Agentic Layers.

4. The cases

  • Meta — Muse for Small Business: launched within weeks of the consumer agent because people were already using it for work; adds business-account and data connections (Shopify, QuickBooks). Meta's instinct: "we're going to add data." Personal → business is a short path.
  • Salesforce — Koa + the Claude partnership: Koa, a CRM-tuned reasoning model (on a "Neotron" stack — almost certainly NVIDIA Nemotron) announced at Dreamforce on 15 September, in pilot, trained on synthetic data informed by 20 years of real CRM flows. Positioned across three levels: data layer, an agent gateway / control panel for anybody's agents, and its own agent. The Claude tie-up is about selling optionality: "every single business leader doesn't want to be locked in… as a seller, you have to sell optionality." Forward-deployed engineers from Anthropic come with it.
  • Microsoft — Copilot and Autopilot: same dilemma, bigger. Data layer (Azure), control plane for agents, the productivity suite nobody dares abandon, and "AI that nobody ever got fired for buying." Autopilot is a persistent agent shipped in preview. The pitch is bricks stacked over time, monetising trust.
  • Labs (OpenAI, Anthropic): "They don't have your data. They don't have existing distribution." Their play is "bet on us for intelligence and for applying it to your data" — which is why they partner with the incumbents who do.
  • DoorDash — text ordering + MCP (30 September): wants your order whether it comes through its assistant or anyone else's agent. "Losing that moment in the interface doesn't mean losing the customer transaction." But: "Being reachable by an agent… is table stakes. It does not establish why someone ought to keep paying you."
  • Crustdata + The Firm: a two-person recruiting business built a custom Claude skill on Crustdata's APIs instead of buying a packaged recruiting tool — keep buying the data, build the thin workflow. A data supplier can move up the stack by helping customers build.
  • Workday payroll agent: "Getting an AI to explain a paycheck is a very different proposition from replacing the system that gets people paid." Configured rules, approvals and dependable execution stay; the agent makes the system more valuable by helping people through complexity. No startup wants to hand-code payroll compliance.

5. What to do

  • Employees: "Talk fluently about the way you use AI" — data, workflows, agent. A hoarded trick is brittle when the vendor changes.
  • Leaders / buyers: understand the fingerprint usage of each team before consolidating; don't "slap AI over the top." Decide where the marginal engineer goes — back-office workflows or competitive advantage.
  • Sellers: sell the 2027 answer — plug-and-play workflows, accessible data, multiplayer agents. "Businesses increasingly are buying outcomes… They are not looking for you to adopt AI. That's the baseline assumption."
  • Closing line: agents "are forcing us to vertically integrate [the] stack… everyone is now a decision maker."

Connects to your work

Contradictions / tensions

  • The vault's Build vs Buy (Agents) page (Praveen Akkiraju, Boris Cherny) says don't rebuild the system of record; Jones agrees on payroll but the Crustdata case shows a point-solution SaaS being displaced by data-plus-skill — the line to defend is the compliance workflow, not every application.
  • Jones's "the data layer gets more valuable" sits uneasily with SaaSpocalypse's YC view that the legacy-code moat is gone; both can be true if the moat was never the code.

Cross-references

Sources

  • Video — YouTube; AI News & Strategy Daily | Nate B Jones; clipped 2026-10-24
  • Raw transcript — Obsidian Web Clipper capture with timestamps