Sunil Venkataram
Back to garden
•
🌱 budding
System of Action - Evolution Beyond System of Record
type/notesaasaiproduct

System of Action — Evolution Beyond System of Record

The three generations

1. Systems of record (1980s-2000s). Databases, ERPs, CRMs, EHRs, PMSs. Their job is to maintain the truth about what happened — customers, transactions, patients, inventory. Value comes from being the authoritative source that other systems have to accept as the ground truth. Oracle, SAP, Salesforce in its early years, Epic, Cerner.

2. Systems of engagement (2000s-2020s). Applications that let humans interact with the record efficiently — workflow software, user-facing dashboards, email marketing tools, scheduling systems. These systems sit on top of the record and their value comes from making the record useful for humans doing jobs. Zendesk, Marketo, Asana, Monday. They sit atop the record and optimize access to it.

3. Systems of action (2020s onward). Platforms where work is decided and triggered — decided and triggered. A system of action can route incoming data, invoke downstream workflows, make judgment calls, and produce output that feeds back into other systems. It’s the place where the loop closes between information and action. A system of action can be used by humans, by AI-assisted humans, or by fully autonomous agents acting on behalf of humans.

The important point is that the system of action is a different architectural layer from the system of record. It can sit on top of the record, or it can route around the record, or it can subsume the record entirely. Whichever happens, the system of action becomes the thing customers actually use day-to-day — and the system of record becomes a downstream dependency rather than a primary interface.

Why the system of action captures value

Data gravity and workflow gravity both migrate to wherever decisions and actions happen. Three specific consequences:

1. The system of action gets first access to the data that matters. When a workflow is triggered somewhere, the trigger point has complete context about why it was triggered. Anything downstream sees a subset. In veterinary practice management, when a transcription tool captures a vet’s session notes, it has the full context. The billing system that ultimately charges for the visit sees only what the transcription tool chose to pass along. Control of the trigger point becomes control of the data.

2. The system of action sets the integration terms. Once a native AI player has enough users in a workflow, they can compel the incumbent system of record to integrate — not the reverse. The Chrome extension becomes an RPA bot becomes a formal API integration becomes a preferred partner relationship becomes a “wait, who’s the incumbent now?” situation. Each step is a small decision and the cumulative effect is a flipped control relationship.

3. The system of action captures the pricing power. Customers pay for the thing they use most, not the thing that holds their oldest data. Once a vet spends 80% of their software time in a transcription + workflow tool and only logs into the old PMS for billing reconciliation, their perception of which software is “valuable” has inverted. The budget follows the perception.

The system of record still matters as authoritative ground truth for audit, compliance, and institutional memory. But leverage moves to wherever decisions and actions happen.

Who wins the race to the system of action

Two candidate winners exist in any vertical:

The incumbent control point. Has data gravity, workflow gravity, existing customer relationships, and brand trust. Can, in theory, build AI features that make their existing platform the natural system of action. Often fails because the instincts of a mature software company are wrong for the transition: they gold plate (perfect before ship), resist low price points, distrust PLG, and prioritize demo-request funnels over self-serve adoption.

The native AI challenger. Starts with zero data gravity, zero customer relationships, zero brand. But has a product that solves a painful problem for a specific user population (usually the Hero users in the vertical) in a way that feels magical. Starts as “just a wrapper” around foundation models with vertical-specific prompts and UI. Often wins because the user adoption compounds faster than the incumbent can respond.

The race structure:

  1. Native AI builds a wedge product that solves one painful problem
  2. Native AI acquires users through PLG, usually targeting Hero users who have tool-selection agency
  3. Native AI uses scrappy integration tactics (Chrome extensions, RPA, virtual users) to connect to the incumbent’s system of record
  4. Native AI reaches enough scale that the incumbent is forced to formally integrate
  5. Native AI expands product surface area to cover adjacent workflows
  6. Native AI’s product becomes the primary interface customers use
  7. System of action has migrated. System of record is now a downstream dependency.

If the incumbent is slow at any step, the native AI wins. If the incumbent ships an equivalent wedge product early, locks down integrations, and shifts to PLG, they can hold the position. Historically most incumbents are slow.

What makes this different from earlier waves

Every previous wave of enterprise software featured “system X will be disrupted by system Y” narratives that sometimes came true and sometimes didn’t. What’s different with the system-of-action concept in 2025-2026:

  • The wedge product is categorically better, not marginally better. A transcription tool that saves a vet 3 hours per day isn’t 20% better than the existing workflow — it’s transformationally better. That changes user adoption dynamics entirely.
  • AI-generated software gets cheaper very fast. The native AI challenger’s product improves on the same cost curve as foundation models. The incumbent’s product improves at the pace of their internal engineering org. The gap widens over time.
  • PLG is mature. The playbook for getting a product into users’ hands without going through procurement is well-understood in 2026, not experimental. Every step of the integrate-and-surround strategy has been run by other companies already.
  • Control-point incumbents are vulnerable in ways they weren’t before. The old defense (“the record is sacred, you can’t replace it”) collapses when the record becomes a downstream dependency of the real interface.

Defensive strategy for incumbents

If you run an incumbent system of record and want to avoid losing the system of action, the playbook is approximately:

  • Identify your Hero users. The practitioners whose experience of your product matters most. The vets in veterinary software, the attorneys in legal software, the doctors in EHR software, the agents in real estate software. The practitioners, not the owners or administrators.
  • Protect the integration points where native AI challengers would attack. APIs, MCP endpoints, RPA-friendly surfaces. Lock them down without being so closed that customers revolt. See also Horizontal Platform to Vertical Specialization - Enterprise Credibility Pivot for why vertical lockdown is the defensible move.
  • Ship wedge products for your own Hero users fast. Stop gold-plating. Ship a transcription tool, a scheduling AI, a notes summarizer, even a “just a wrapper” version, before the native AI does. Build for users, not for owners.
  • Adopt PLG motion even if your sales force hates it. Free trials, instant sign-up, instant value. Your sales team will fight this; it will still be the right answer.
  • Reconsider pricing. Move from seat-based to consumption or outcome pricing where possible. See Seat-Based to Token-Based SaaS Pricing Transition. Seat pricing is the enemy of PLG and is also structurally short the AI transition.
  • Ship good now; perfect later. Ship today, iterate, beat the native AI to the wedge. Your existing data and workflow advantages mean nothing if the native AI gets there first.

Counter-argument — distribution is still a moat

A serious pushback on the universal form of this thesis, held as a live tension:

Balaji Srinivasan argues in a 2026 interview that the “incumbents are dead” framing confuses code with distribution. Native AI challengers can clone the code cheaply, but the accumulated user relationship is a separate asset that cannot be replicated in the same way. AI accelerates both incumbents and disruptors equally, which preserves the relative gap in percentage terms even as both sides speed up in absolute terms. See Distribution as the Remaining Moat - Why SaaS Incumbents Aren’t Dead.

The reconciliation that holds both positions honestly: split incumbents into two categories.

  • Category A — healthy incumbents with real distribution. Product still improving, user base growing on merit, engineering culture can absorb AI. For this category, the system-of-action argument overstates the threat. Balaji’s distribution-moat holds. Examples include Figma, Linear, Shopify in its main verticals.
  • Category B — complacent incumbents milking installed bases. Product stagnant, users retained by switching costs rather than by value delivered, engineering culture resists shipping. For this category, the system-of-action argument holds — they WILL get flipped by native AI challengers. Balaji explicitly names NetSuite in this bucket.

The universal form of “native AI flips all incumbents” is overstated. The bifurcated form — “native AI flips Category B incumbents; Category A incumbents hold if they move fast enough” — survives both serious arguments. The diagnostic question for any specific company is which category it belongs to, and the answer is usually clear if you look at engineering velocity, user growth source, and product-quality trajectory over the last 12 months.

Where the system-of-action frame cuts across other theses

This framing connects cleanly to several other live arguments:

  • Claws - Persistent Looping Agents as App Replacement (Karpathy): “apps should be APIs, agents should be the glue, the customer is increasingly an agent.” Same observation from a consumer angle. System of action = where the agent’s glue lives.
  • Seat-Based to Token-Based SaaS Pricing Transition (David George): token pricing and the “agent-can-pay test.” System of action captures pricing power because it captures the workflow the agent actually runs through.
  • Horizontal Platform to Vertical Specialization - Enterprise Credibility Pivot (Luminai/Kesava): vertical wedges win enterprise credibility. Hero User strategy is the GTM mechanism for how a vertical wedge grows.
  • Healthcare Admin Automation - Data Transformation and Vertical Workflows: same two-layer architecture (data transformation + verticalized workflow agents) described in the healthcare context. System of action is the abstraction that generalizes it.

Four independent sources (Karpathy, David George, Kesava Kirupa/Luminai, Tidemark) all arriving at variations of “the agent/workflow layer captures the value, not the record layer.” That’s an unusually strong convergence signal.

Backlinks (7)
Connected Notes