Systems of Action: The New AI Layer Above Systems of Record
.webp)
.webp)
.webp)
Why the next trillion-dollar AI category won’t be owned by Salesforce, Workday, or integration tools, and why Cursor is the first real proof point
A system of action is an AI-native layer that sits above systems of record and turns intent into execution across tools.
Systems of record (CRM, ERP, ticketing, HRIS) are still essential. They store authoritative data, enforce permissions, and support auditability. But they were never designed to do the thing people actually struggle with every day: coordinate work across tools, handle exceptions, and complete outcomes.
That’s the opening for a new category: the system of action.
A system of action doesn’t just answer questions. It:
- interprets intent (“what we’re trying to get done”),
- orchestrates steps across tools and teams,
- coordinates humans + software,
- and learns from real-world resolution patterns (approvals, overrides, edge cases).
This layer is the next major shift in enterprise software, and the most important part is what it cannot be: it cannot be owned by a single system of record vendor, and it is unlikely to be owned by traditional integration tooling.
Systems of Action, explained simply
- What it is: An AI layer that owns intent, decisions, and execution
- Where it lives: Above systems of record like Salesforce, SAP, Zendesk, Workday, Jira
- What it replaces: Manual coordination, tribal knowledge, brittle workflows, swivel-chair operations
- What it enables: Faster execution, fewer handoffs, continuous learning from outcomes
- Why it matters: Value shifts from storing data to making things happen
Why “systems of action” are suddenly everywhere
Over the last year, a clear pattern has emerged across venture and operator analysis:
- Foundation Capital: Context graphs and decision traces
- Menlo Ventures: State of generative AI in the enterprise
- Clouded Judgement (Jamin Ball): Long live systems of record
Despite different language, these pieces converge on a shared conclusion:
AI is creating a new software layer that sits above systems of record and turns intent into execution.
That layer is increasingly described as a system of action.
What is a system of action?
A system of action is not:
- a database,
- a reporting layer,
- a workflow engine,
- or an “AI copilot” bolted onto an existing product.
A system of action is a layer that:
- Captures intent
Users express goals in natural language (“fix this customer escalation,” “renew this account,” “resolve this incident”), not in object updates. - Orchestrates multi-step execution
It decomposes goals into steps across multiple systems of record (and sometimes across humans). - Coordinates humans and software
It knows when to auto-execute and when to route to a person for approval, input, or exception handling. - Handles exceptions as a first-class concept
Enterprise work is mostly exceptions. Systems of action learn the real “how work gets done” path, not the idealized one. - Improves via decision traces
It gets better over time by observing outcomes, approvals, overrides, and resolution patterns, not just by adding more rules.
In one line:
Systems of record store what is. Systems of action decide and execute what should happen next.
| Layer | Primary Role | What It Owns | What It Doesn’t Own |
|---|---|---|---|
| System of Record | Store truth | Data models, permissions, compliance | Intent, orchestration |
| Integration Layer | Move data | Triggers, APIs, pipelines | Decisions, learning |
| System of Action | Execute work | Intent, workflows, decision traces | Underlying data |
Why systems of record can’t own this layer
The core argument in the sources is not that systems of record are dying, it’s that they are insufficient as the control layer for AI-era execution.
1) Economic conflict
Systems of record are incentivized to deepen lock-in to their own models and workflows. A system of action must orchestrate across multiple systems, including competitors, which creates an inherent conflict.
2) Architectural limits
Systems of record are:
- object-centric,
- schema-driven,
- optimized for accuracy and compliance.
Systems of action are:
- intent-centric,
- context-driven,
- optimized for speed, coordination, and outcome completion.
3) Ownership mismatch
Systems of record are owned by IT/admin/platform teams.
Systems of action are owned by operators:
- support leaders,
- revenue teams,
- finance ops,
- process owners.
That mismatch matters because whoever owns the system defines what “success” means. Systems of action are evaluated on outcomes, not data cleanliness.
Why integration platforms also don’t fit
It’s tempting to think the system-of-action winner will be an integration vendor, because integration touches many systems. But classic integration platforms primarily:
- move data,
- trigger workflows,
- connect APIs.
They typically do not:
- own intent,
- accumulate durable context across time,
- capture decision traces,
- learn from outcomes in a closed loop.
They are critical plumbing, but not the decision-and-execution layer.
A system of action may use iPaaS. It won’t be iPaaS.
What this new layer actually owns
Across the sources, the system of action consistently owns five things:
- Intent: what the user/org is trying to achieve
- Orchestration: which systems, steps, and people are involved
- Exceptions: what happens when reality diverges from the happy path
- Decision traces: who decided what, when, and why
- Execution: completing outcomes, not just recommending next steps
This is why Foundation emphasizes context graphs: you need a living map of relationships, history, and decisions to execute reliably across systems and time.
Why this layer barely exists yet
If the opportunity is so big, why are there so few real examples?
Because building a system of action requires something most vendors don’t have:
Control of the action surface where real work happens.
Without that surface, you can’t:
- capture intent naturally,
- observe decision-making,
- close the loop with outcomes,
- or improve continuously.
Most “AI agents” today live as add-ons, useful, but not the execution layer.
Cursor: the first real system of action
This is why Cursor stands out as the cleanest proof point.
Cursor:
- does not replace GitHub, Jira, or CI tools,
- does not own code as a system of record,
- sits where work happens: the IDE,
- turns intent into multi-step execution,
- learns from retries, edits, and accept/reject loops,
- improves through usage, not configuration.
Cursor doesn’t just help you get an answer.
It owns the execution loop of software development.
That is what a system of action looks like when it’s real.
Why Cursor is (so far) the only clear example
Software development uniquely offers:
- a dominant action surface (the IDE),
- clear atomic actions,
- fast feedback loops,
- relatively low compliance friction.
Most enterprise domains (support, revenue, finance, ops) don’t yet have an equivalent execution surface. That’s why the category feels obvious in theory, but thin in concrete examples.
The lack of examples is not a weakness of the thesis.
It’s proof the category is just beginning.
What this means for the future of enterprise AI
The takeaway is not “build better copilots.”
It’s this:
The next generation of enterprise software will be systems of action, AI-native layers above systems of record that turn intent into execution.
Systems of record will remain the source of truth.
Integration will remain essential plumbing.
But value will increasingly move to the layer that decides and acts.
Cursor is the first clear signal of what’s coming next.
Why Worknet is focused here
Worknet is an AI adoption agent for the software your customers use and the software your employees use. It sits with every user inside the app, offering proactive support and onboarding. It answers with context, acts on the user's behalf within the permissions you set, and brings in the right person from your team when needed. For SaaS companies, what it learns becomes expansion signals for sales, risk alerts for success, and fixes for product. For IT, it rolls out across the apps you buy with no vendor changes and shows where employees get stuck. An AI-native alternative to Pendo, WalkMe and Appcues.
At Worknet, we believe the opportunity is not replacing systems of record. It’s building the action layer above them, where real work happens, decisions are made, exceptions are handled, and outcomes are driven end-to-end.
That’s the system of action era.
And it’s just getting started.
References
- Foundation Capital, “AI’s trillion-dollar opportunity: Context graphs” (Dec 22, 2025)
https://foundationcapital.com/context-graphs-ais-trillion-dollar-opportunity/ - Menlo Ventures, “2025: The State of Generative AI in the Enterprise” (Dec 9, 2025)
https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/ - Clouded Judgement (Jamin Ball), “Clouded Judgement 12.12.25: Long Live Systems of Record” (Dec 12, 2025)
https://cloudedjudgement.substack.com/p/clouded-judgement-121225-long-live - Systems of record lose ground - Sarah Wang https://a16z.com/newsletter/big-ideas-2026-part-1
FAQ's
Frequently Asked Questions
What is a system of action?
A system of action is an AI-native layer that turns user intent into execution across tools. It sits above systems of record, orchestrates multi-step work, and improves over time by learning from decisions, exceptions, and outcomes. Source: Foundation Capital
How is a system of action different from a system of record?
A system of record stores authoritative data and supports compliance and auditability. A system of action owns intent, orchestration, and execution across multiple systems of record without replacing them. Source: Clouded Judgement
Why can’t systems of record (Salesforce, SAP, Workday) own the system of action layer?
Because the system-of-action layer must coordinate across many systems and capture decision traces and intent beyond any single vendor’s domain model. Systems of record are optimized to store truth inside their own boundaries, not orchestrate workflows across competitors. Source: Foundation Capital
Why can’t integration platforms (iPaaS) own this layer?
Integration platforms are great at moving data and triggering workflows, but they typically don’t own intent, don’t accumulate decision traces, and don’t learn from outcomes the way an action layer must. The system-of-action layer may use integration tooling, but it’s not the same category. Source: Clouded Judgement
What are “context graphs” and how do they relate to systems of action?
A context graph is a way to represent the relationships between people, objects, decisions, and history across systems. Context graphs help agents operate safely and effectively by grounding actions in real organizational context and decision traces. Source: Foundation Capital
Does generative AI spending suggest this layer is real?
Yes. Enterprise spending has rapidly shifted toward the application layer built on top of foundation models, which supports the emergence of workflow- and action-oriented products above systems of record. Source: Menlo Ventures
Why is Cursor often used as the best example of a system of action?
Cursor is frequently cited as a proof point because it sits above the underlying systems of record (repos, issue trackers, CI) and becomes the execution surface where intent turns into multi-step work. It’s one of the clearest examples of a product that doesn’t just answer questions, it drives outcomes through action loops. (This is an interpretation based on the “action layer” framing in the sources.) Source: Clouded Judgement
Are systems of record dying?
No. A common view is that systems of record remain essential sources of truth, but they are being “rewired” as agents become the new front door and demand more reliable semantics and safe read/write controls. Source: Clouded Judgement
.png)
