The Agentive Product Stack: How the Layers Compose

The taxonomy describes roles. Products of Consequence, Judgment, Counsel, Action. Four categories, four different jobs.

But products don't exist in isolation. They compose into systems. A Product of Counsel consumes signals from Products of Judgment. A Product of Consequence orchestrates the whole stack, embedding Counsel, consuming Judgment, and coordinating Action.

Understanding the dependency structure guides how you build your product. Where you sit, what you depend on, who depends on you.

The Stack

Four layers. Each depends on the layers below.

┌─────────────────────────────────────────────┐
│         Products of Consequence             │
├─────────────────────────────────────────────┤
│           Products of Counsel               │
├─────────────────────────────────────────────┤
│           Products of Judgment              │
├─────────────────────────────────────────────┤
│            Products of Action               │
└─────────────────────────────────────────────┘

Products of Action are the foundation. Idempotent operations. Send the text, process the payment, store the file. They don't reason. They execute.

Products of Judgment sit on top. They consume Action and produce derived signals. Not retrieval. Inference, matching, synthesis, assessment. A fraud score, a match confidence, a forecast. The output includes not just the answer but how much to trust it.

Products of Counsel sit on top of that. They advise humans toward outcomes. They consume Judgment for grounding. You can't give good style advice without sizing data. You can't give good financial advice without market signals. They may invoke Action directly. The accumulated context about the user, combined with domain expertise, produces advice worth taking.

Products of Consequence sit at the top. Agents act, humans supervise, accountability lives here. They embed Counsel as a capability. Advisory modes for the arbiter at decision points. They consume Judgment for the signals agents act on. They coordinate Action for execution. The whole stack serves the outcomes that happen at this layer.

This is a hierarchy of dependencies, not a hierarchy of value. Your product enters where it enters. A style advisor is a Product of Counsel. That's the terminal point. The human takes the advice and acts on their own. A mortgage platform is a Product of Consequence. It orchestrates everything below.

What Flows Up

Each layer provides what the layer above needs to function.

Action provides reliable execution. The operations that everything else is built on. When Judgment needs to check a data source, when Counsel needs to send a recommendation, when Consequence needs to move money, Action executes.

Judgment provides signals. Not raw data. Assessed data. Confidence scores, dimension breakdowns, and reason codes. The derived intelligence that Counsel uses to ground advice and Consequence uses to inform agent behavior.

Counsel provides guidance. For standalone Counsel, that guidance goes directly to humans. For embedded Counsel, it serves the arbiter within a Product of Consequence. Advisory modes at decision points. Helping the human supervisor navigate complexity.

Each layer transforms what it receives from below into something the layer above can use. Action becomes signal. Signal becomes advice. Advice becomes supervised outcomes.

What Flows Down

Requirements.

The doctrine at each layer isn't arbitrary. It's the interface contract demanded by whatever builds on top.

Products of Consequence need decision-ready signals. That requirement flows down to Judgment. Structured confidence, legible reasoning, and clear provenance. An agent can't act well on a naked answer. The arbiter can't supervise what they can't understand. Judgment must provide the structure that makes both possible.

Products of Counsel need grounded expertise. That requirement also flows down to Judgment. Accurate signals, clear about what's derived versus retrieved, transparent about staleness and uncertainty. You can't give good advice on bad information. Counsel needs Judgment to be honest about what it knows and how well it knows it.

Products of Judgment need reliable execution. That requirement flows down to Action. Operations that work consistently at the latency the workflow demands. If Action is flaky, Judgment can't do its job.

The doctrine pieces aren't standalone manifestos. They're specifications. Each one describes what the layers above require from you. Miss those requirements, and you break the products that depend on you.

Where You Enter

Your product enters the stack at one layer. That determines your core job, your doctrine, and your relationships.

Enter at Action. Your infrastructure. Reliability, speed, price. The job is execution. Do the thing, do it well, do it fast. Commodity dynamics apply unless you have something proprietary. Your customers are everyone above you in the stack.

Enter at Judgment. Your signal. Accuracy, confidence structure, and legible reasoning. Your customers are Products of Counsel and Products of Consequence. They need your output to be decision-ready, not raw. The interface contracts are demanding because the layers above can't function without structured uncertainty. Get it right, and you're critical infrastructure. Get it wrong, and you're replaced by someone who takes the requirements seriously.

Enter at Counsel. You're advisory. Accumulated context, encoded expertise, trust through honesty. Some Products of Counsel are standalone. The human is the arbiter of their own life, and your advice helps them navigate it. Style, coaching, health, finance, learning. The terminal point is the human acting elsewhere. Other Products of Counsel get embedded into Products of Consequence. Advisory modes for arbiters at decision points. Different go-to-market, same doctrine. The moat is accumulated user context and domain judgment that competitors can't replicate.

Enter at Consequence. You're orchestration. Agents act, humans supervise, you're accountable for outcomes. You consume from every layer below. You likely embed Counsel as a capability for your arbiters. You definitely consume Judgment for the signals your agents act on. You coordinate Action for execution. The moat is the supervision surface, the trust gradient mechanics, and the workflow itself. The doctrine is the most demanding because the stakes are highest.

Where the Moats Are

Different at each layer.

Action. Reliability and price. Moats are hard to build here. Proprietary infrastructure, regulatory capture, and network effects, if you can get them. Otherwise, commodity.

Judgment. Accuracy and confidence structure. The moat is being right more often than alternatives and being useful about uncertainty when you're not sure. Proprietary data sources, superior models, and investment in the structured confidence that the layers above require. If your signals are better and better-explained, you win.

Counsel. Accumulated context and encoded expertise. The moat is knowing this user. Their situation, their history, their constraints. A new entrant can't replicate that. Combined with domain judgment that takes investment to build. The product that's been advising someone for two years has context that a competitor starting today can't match.

Consequence. The supervision surface and trust mechanics. The moat is the workflow. The orchestration, the visibility, the intervention capabilities, and the trust gradient that lets arbiters grant autonomy selectively. Hard to build, hard to replicate. Products of Consequence that get the arbiter model right create lock-in through learned trust patterns. Users grant autonomy to agents they've calibrated. Switching means starting the trust gradient over.

Competing at the wrong layer with the wrong moat is a strategy error. A Product of Judgment trying to win on price competes with commodities. A Product of Counsel without accumulated context competes with ChatGPT. A Product of Consequence without supervision, mechanics competes with fear.

Architecting With the Stack

This is a thinking tool. When you're designing a product for the agentive era, the stack gives you the domains to reason about.

What layer are you entering? That tells you your core job and points you to your doctrine. If you're building Counsel, the Counsel doctrine applies. If you're building Consequence, you need it all.

What do you depend on? Those are your integration decisions. Not just build or buy. What interface contracts do you need from the layers below? If you're building Counsel, what does your Judgment layer need to provide for your advice to be grounded? If you're building Consequence, what signals do your agents need to act well?

Who depends on you? Those are your customers' requirements flowing back. If Products of Consequence will consume your Judgment signals, your confidence structures aren't optional. They're the interface contract. If your Counsel will be embedded in someone's workflow, your advisory modes need to fit their supervision surface.

Where are the seams? The boundaries between layers are where integration happens. Also, where leverage exists. A Product of Counsel that builds proprietary Judgment has a moat that competitors can't replicate by plugging into the same APIs. A Product of Consequence that builds proprietary Counsel creates advisory capabilities tuned to its specific workflow. The seams are strategic choices.

What's your build/integrate decision at each dependency? You can build your own Judgment layer or consume external APIs. You can build advisory capabilities or integrate specialized Counsel. The stack doesn't dictate the answer. It clarifies what you're deciding about. Each dependency involves trade-offs among cost, control, and differentiation.

The Landscape

The agentive era reorganizes software into layers. Action at the foundation, Judgment on top of it, Counsel on top of that, Consequence at the apex. Dependencies flow up. Requirements flow down.

Every product finds its place in this stack. Some are standalone at their layer. A Product of Counsel that advises humans directly. Terminal point, complete in itself. Some orchestrate the layers below. A Product of Consequence that embeds Counsel, consumes Judgment, coordinates Action.

The products that thrive understand the stack. They know their position in it. They know what they depend on and what depends on them. They build deliberately for the layers above and below.

The products that struggle fight the structure. They build Judgment without confidence signals because it's hard. They build Counsel without accumulated context because it's slow. They build Consequence without supervision because they trust the automation. Each failure mode is a refusal to meet the requirements the stack imposes.

The stack isn't optional. It's the shape of software in the agentive age. Build for it.

This article is the capstone of the GenAI Product Doctrine series.

Start with the taxonomy to find where your product fits: A Product Taxonomy for the Agentive Age →

Then read the doctrine for your layer:

Products of Consequence → — Where outcomes happen, agents act, and humans arbitrate.

Products of Judgment → — Your output is someone else's input. Make it decision-ready.

Products of Counsel → — Advice that advances outcomes. The moat is being worth believing.