Data, Application and Agentic Layers

Nate B. Jones's three-layer frame for reasoning about software once agents do the information-processing part of a role. Every employee, buyer and vendor now has to be able to say where they stand against each layer:

Layer	What it is	What agents do to it
Data	Records, structured information, context nobody else has (candidate data, customer records, configured payroll rules)	Agents need it desperately; whoever supplies it keeps getting paid. The layer gets more valuable.
Application	Decomposes into interfaces (the screens humans click) and workflows (structured, often compliance-bearing sequences on top of data)	Interfaces get thinner or bypassed; workflows people don't want to touch (payroll, approvals) stay valuable. "The application layer is dead" is wrong — it is being decomposed.
Agentic	The agent, its harness, memory and the operator's accumulated instructions	Absorbs the "finding, arranging, acting" work SaaS used to bundle; becomes the new sticky layer (Agent Lock-In).

The diagnostic question that drives it: "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 of finding, arranging, acting on and storing information; an agent getting good at part of that exposes the whole relationship, because the vendor never said which part it was really being paid for.

How the named players map

Data + distribution incumbents (Salesforce, Microsoft) reason from the data they hold into the agent position they think they can occupy: data layer → agent gateway/control plane → own agent (Koa, Copilot/Autopilot). Their pitch becomes a multi-value stack rather than a single demo.
Labs (OpenAI, Anthropic) have neither your data nor distribution; they sell intelligence and the ability to apply it — hence partnerships that let incumbents sell optionality (Salesforce + Claude).
Consumer agents moving into business (Meta's Muse for Small Business) start at the agentic layer and bolt on data connections (Shopify, QuickBooks).
Be-reachable players (DoorDash text ordering + MCP) accept losing the interface moment to keep the transaction. "Being reachable by an agent… is table stakes. It does not establish why someone ought to keep paying you."
Thin-workflow builders (Crustdata + The Firm) keep buying the data layer and build the workflow themselves — see Build vs Buy (Agents).
Compliance-workflow owners (Workday payroll agent) show the application layer surviving as workflow, not interface.

What each seat should do (Jones)

Employee: talk fluently about how you use AI — which data, which workflows, which agent; a hoarded "special trick" is brittle the day the CFO switches vendors.
Leader / buyer: audit the fingerprint usage team by team before consolidating; "do not sit at the board level and say we understand AI… and slap AI over the top." Decide where the marginal engineer goes: back-office workflows or competitive advantage.
Seller: sell the 2027 answer — plug-and-play workflows, an accessible data layer, agents that are multiplayer (yours is great at X; others can come in). Buyers are buying a belief you will stay current when the next model lands.

Vault connections

Sharpens SaaSpocalypse: not "SaaS dies" but "SaaS value decomposes by layer" — the seat was pricing the interface; the data and the compliance workflow were the value.
Build vs Buy (Agents) gets a layer-aware rule: buy data and compliance workflows; build the agentic layer and thin workflows on top.
Agentic Commerce's "dual front door" is the consumer-side twin of DoorDash's reachable-by-any-agent posture.
Harness (LLM Agents) and Agent Memory are what the agentic layer is made of — and why it locks in.

Sources

The End of the App Era What Comes Next (Nate B Jones) — 2026-10; canonical statement (~24:30), cases throughout