When Camunda — a 700-person BPM company most operators have never heard of — announced ProcessOS today at CamundaCon, the framing was unusually specific. They didn't call it "an AI feature." They called it an "intelligence layer." A distinct stratum of the stack. Something that sits above the workflows and decides how they're going to run.

That phrasing matters. For two years, agents have lived inside applications — a chat sidebar here, a Slack bot there. With ProcessOS, Camunda is the latest of a half-dozen vendors to declare that agents have their own home, their own SLA, their own pricing model, their own org chart. SAP did this in February with Joule. Salesforce did it in November with Agentforce. Now Camunda. The agent layer is moving from metaphor to product category.

For indie operators selling agents into businesses, this changes what you're really selling — and how it fits into the architecture the buyer is now planning around.

What "Agent Layer" Actually Means

Strip the marketing and an agent layer is three things stacked: a registry, an orchestrator, and a governance plane. Camunda's ProcessOS hits all three.

Registry. A catalog of every agent the company runs — who built it, what it does, what data it can touch, what it costs, who owns it. Without a registry, the CFO has no idea how much the agent stack costs each month. With a registry, every agent is a line item.

Orchestrator. The thing that decides which agent runs when, in what order, with what data. ProcessOS specifically markets the ability to "re-engineer business processes as agentic workflows" — meaning the BPM diagram you used to draw in Visio is now a runnable, agent-aware graph.

Governance plane. Logs, audits, retries, kill-switches. The boring infrastructure that lets a regulated enterprise actually deploy agents without their compliance team filing a lawsuit. ProcessOS, Joule, and Agentforce all include this — because without it, no FT500 buys.

For the indie operator, this means your agent isn't being deployed into a vacuum. It's being deployed into a layer that already has opinions about how agents should behave.

How This Changes What You Sell

Most indie agents shipped in the last 12 months were "applications" — standalone tools with their own UI, their own auth, their own dashboard. The customer ran them in isolation. The agent layer changes that.

Your agent becomes a node. A buyer running ProcessOS or Agentforce wants to register your agent inside their orchestrator, not maintain a separate login for it. If your agent doesn't expose a clean callable interface (HTTP, MCP, gRPC), it doesn't fit. Adapt now.

Your dashboard becomes optional. The buyer's intelligence layer is the dashboard. Your agent is the worker. Cling to your beautiful dashboard and you'll lose deals; expose the data instead and you'll get adopted by every customer who has an agent layer.

Your pricing has to pass orchestrator-level scrutiny. Once your agent is a line item in a registry next to twenty others, the CFO compares costs. Per-resolution agents at €1.50 look obvious. Per-seat agents look strange. Update accordingly.

Your contract terms have to be light. Heavy contracts that took weeks to negotiate die in the orchestrator era. The new model: month-to-month, API-key auth, an SLA in a markdown file. The lighter the contract, the easier the registration, the higher the adoption.

Where This Lands by Buyer Type

The agent layer rollout isn't going to be uniform. Different buyer segments will move at different speeds, and the indie operator's GTM should match.

Large enterprise (5,000+ employees). Camunda, Salesforce, SAP, Microsoft, ServiceNow — at least one of these is running on their stack. The agent layer is real and your agent has to integrate. Sell to the agent-layer admin, not the line-of-business buyer.

Mid-market (200–5,000 employees). Just discovered the agent layer concept. Will pick one in the next 12 months. Your opportunity is to ship before they lock in, become the agent they registered, and ride the orchestrator-led adoption inside the company.

SMB (under 200 employees). No agent layer. No registry. No orchestrator. The standalone application model still works perfectly here — and will for years. This is the most attractive segment for solo operators specifically because the layer-thinking doesn't apply.

Solo / micro-business. Same as SMB but more so. They want a Zapier-style "it just works" experience. Don't even mention orchestration.

What Indie Operators Should Do This Week

The Camunda news isn't urgent for most builders — but it's a useful forcing function. Three concrete moves to make.

Add a clean HTTP endpoint to your agent. Whatever form it takes today (Slack bot, web app, browser extension), expose the core capability as a single HTTP POST. This is the unit of integration with every agent layer being built. Two hours of work; permanent optionality.

Publish your "agent spec." A two-page markdown file: what your agent does, what inputs it accepts, what outputs it produces, what it costs, what it doesn't do. Buyers running orchestrators ask for this. Most indie agents don't have one. Having it is a competitive advantage.

Pick which agent layer to integrate with first. You can't integrate with all of them. Look at where your customers are. Most B2B SaaS deals will run through Agentforce. Most regulated-industry deals will run through ProcessOS or ServiceNow. Manufacturing will run through SAP Joule. Pick one, ship the integration, and write a case study.

Keep selling the standalone version. SMB and solo buyers still want a one-click experience. The agent-layer story is true for enterprise but false for everyone else. Don't kill your highest-margin segment chasing the trend.

The Real Story Beneath the Beta

ProcessOS, Joule, Agentforce, and the orchestration products that will land in the next six months are all betting the same thing: that agents will be governed as a class, not deployed as features. That bet is almost certainly right at the enterprise tier and almost certainly wrong at the SMB tier.

Indie operators get to choose which world they live in. Some of you will ship integrations and become a cell on a Camunda dashboard at twenty enterprises. Some of you will sell €390/month agents directly to dentists and never touch an orchestrator in your life. Both are real businesses. Neither is better. The only mistake is to try to do both at once.

The agent layer is real. It's also optional. Pick deliberately.