GenAI Product Doctrine: Products of Counsel

Your job is to help someone get somewhere. Not to the answer. To the outcome.

A Product of Counsel advises humans directly. Recommendations, guidance, expertise on demand. The outcome happens elsewhere: an outfit gets worn, a purchase gets made, a decision gets executed. The conversation serves something beyond itself.

This excludes companionship products, where the conversation is the point. Those are real, they’ll be huge, and they’re out of scope here.

The Failure Mode

Everyone has a chatbot now.

Take your product’s data, throw it into context, put a chat interface on top, ship it. “We have AI.” Every product team in the world is doing this.

It’s table stakes. There’s no moat.

The interface is commoditized. The models are commoditized. If your product is “chat with our data,” you’re competing on data alone. Most data isn’t defensible enough to build a business on.

This is how most teams approach Products of Counsel. They’ve built a retrieval interface with a conversational skin. That’s not counsel. That’s search with extra steps.

What Counsel Actually Requires

You can’t give good advice without knowing what you’re advising about.

A style advisor without your closet, your body, your preferences, your recent outfits? That’s generic fashion tips. A leadership advisor without your org, your challenges, your history? That’s management platitudes. The advice is only as good as the context it’s grounded in.

The failure mode throws the product’s data at the model. Real counsel requires deep knowledge of the person being advised. Their situation. Their goals. Their constraints. Their history. Without that, you’re giving the same advice to everyone. That’s not counsel. That’s content.

Where the Moat Lives

The moat is in the data you accumulate. Two kinds.

User context is everything about this person that makes the advice specific. Their closet. Their org structure. Their patterns. What they’ve tried, what worked, what didn’t. This compounds through use. Every interaction adds signal. A new competitor has the same models you do. They don’t have the relationship you’ve built.

Domain judgment is derived from data about the domain itself. A style advisor needs sizing intelligence: how brands run, how fits compare, what “medium” means across manufacturers. That’s not retrieval. That’s judgment built through investment. You go get data others don’t have and transform it into something that makes advice accurate.

User context without domain judgment gives personalized mediocrity. Domain judgment without user context gives generic expertise. You need both.

Expertise Is in the Reasoning

Data isn’t enough. The question is whether your product reasons about that data the way an expert would.

A good style advisor doesn’t just know what’s in your closet. It has taste. It understands proportion, color, and occasion. It reasons the way a skilled stylist would.

A good leadership advisor doesn’t just know your org chart. It understands political dynamics, timing, when to push, and when to wait. It reasons the way a seasoned executive would.

This expertise lives in prompting. How you’ve encoded domain knowledge, reasoning patterns, and the heuristics a specialist applies. We’re still figuring out how to do this well. But that’s where differentiation lives. Anyone can stuff data into context. Not everyone can encode expert reasoning.

Sycophancy Is the Existential Threat

Every LLM wants to tell you your work is great.

This kills Products of Counsel slowly. It feels good, but it quietly erodes trust.

If your product tells everyone their outfit looks good, it’s useless. If it tells every executive their strategy is sound, it’s dangerous. The human learns that the praise is meaningless. They’ll use you for drafts, for information, for a first pass. They’ll go elsewhere for real judgment.

The products that win in Counsel are the ones humans believe. When it says “that works,” you trust it. Because it was willing to say “that doesn’t work.” The hard truths make the praise meaningful.

Optimizing for user satisfaction means being positive. Right? No. That’s optimizing for short-term comfort at the cost of long-term trust. Trust is what makes people come back.

False Confidence Kills You Too

The other trust-killer is overreach. Advice delivered with confidence on insufficient context.

An expert knows the limits of their expertise. A good counselor says, “I can’t advise you on this yet” when they don’t have what they need. A product that guesses when it lacks context breaks trust just as fast as one that flatters.

This connects back to the context problem. If the product doesn’t know your risk tolerance, your constraints, your history, it should say so. “I don’t have enough to advise you on this” is a better answer than a confident guess that turns out wrong.

The failure mode is a product that always has an answer. Real expertise includes knowing when you don’t know enough.

Trust Unlocks Stickiness

In Products of Consequence, trust unlocks autonomy. The agent earns the right to act without approval. The gradient runs from “approve everything” to “just do it.”

In Products of Counsel, trust unlocks continued use. The product earns the right to be your go-to. The gradient runs from “let me try this” to “this is where I go for this kind of thinking.”

You’re not competing on leverage within a workflow. You’re competing to be the trusted advisor for a category of decisions.

You can’t shortcut this. The products that earn it do so through consistent quality over time, including when that means delivering hard truths. Reliability breeds loyalty.

Counsel Standalone and Counsel Embedded

Products of Counsel take two forms.

Standalone Counsel is a product in its own right. A style advisor. A leadership sounding board. A writing partner. The human comes for advice, gets it, and executes elsewhere.

Embedded Counsel is a mode within a Product of Consequence. The human is supervising agents, hits a decision point, and shifts into a counseling interaction. They work through the problem, reach clarity, and return to supervision.

The form factor differs. The doctrine applies to both.

The Doctrine

Counsel is instrumental. The conversation serves an outcome that happens elsewhere. If you’re not advancing someone toward something, you’re providing entertainment or companionship: different products, different rules.

Context is mandatory. You can’t give good advice without knowing what you’re advising about. Accumulating user context is a product problem: integrations, memory, structure. Solve it or stay generic.

The moat is accumulated data. User context compounds through use. Domain judgment compounds through investment. Both create defensibility that a new entrant can’t replicate by plugging into the same models.

Expertise lives in reasoning. Encode how a specialist thinks, not just what they know. This is where differentiation lives.

Sycophancy kills you. Hard truths build trust. Flattery destroys credibility. Optimize for being believed, not being liked.

False confidence kills you, too. An expert knows the limits of their expertise. If you lack the context to advise, say so. A confident guess that turns out wrong breaks trust faster than admitting you don’t know enough yet.

Trust unlocks stickiness. The goal isn’t autonomy. It’s becoming the go-to. Earned through reliability over time.

The Opportunity

Chat interfaces are everywhere. Data is everywhere. The products that win in Counsel aren’t the ones with the best chat UI or the most data.

They’re the ones that accumulate context about you. That builds domain judgment that others don’t have. That encodes expertise into reasoning. That tells you the hard truths.

The ones that earn trust by being worth trusting.

Counsel is instrumental. Help someone get somewhere. That’s the job.