06-reference / concepts

products for agents

Sat Apr 04 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·concept ·status: active

Products for Agents

The emerging category of products designed for consumption by AI agents, not (only) human eyeballs. As agents become the primary interface for knowledge work, the products that serve them — structured knowledge, machine-readable context, real-time data feeds — become the new infrastructure layer.

The Shift

Yesterday: products for humans (dashboards, reports, articles, apps) Today: products for humans + agents (dual-format, structured + readable) Tomorrow: products primarily for agents (APIs, context graphs, knowledge packs)

Most of today’s enterprise data is locked in formats optimized for human consumption — PDFs, dashboards, slide decks, wiki pages. Agents can parse these, but poorly. The opportunity is in building data products that are native to agent consumption: structured, semantic, queryable, and composable.

Where This Shows Up in Our Work

Data Dots (product concept) — human-readable flashcards that double as an agent-ingestible knowledge base. The novel positioning: “download this to give your AI context about data engineering.” This is Karpathy’s idea file pattern applied as a product.

The vault itself — our Obsidian vault indexed by QMD is a product-for-agents that we built for ourselves. The pattern (structured markdown + semantic search + wikilinks) is productizable.

Data Marketplace (project) — curated industry datasets with semantic context, not just raw data. The value isn’t the data itself — it’s the context that makes the data useful to an agent. This connects to the Foundation Capital thesis that decision traces are the next trillion-dollar platform.

Sanity Check newsletter — articles structured as both human-readable AND agent-ingestible (frontmatter, wikilinks, structured sections). The Nicolas Cole approach of writing for platforms, adapted: write for humans AND agents.

phData consulting — helping enterprises make their internal data “agent-ready.” The context fragmentation problem from Ivan Zhao’s essay: agents can’t work because context is scattered across dozens of tools. The consulting play is consolidating that context into agent-consumable formats.

The Underlying Pattern

Every product-for-agents needs three things:

  1. Structured content — machine-parseable format (JSON, markdown with frontmatter, APIs). Not PDFs, not dashboards, not slide decks.
  2. Semantic context — not just data, but what the data means, how it relates to other data, and when to use it. This is what event clocks and knowledge graphs provide.
  3. Composability — the product should be a building block that agents can combine with other products. This is the skills-as-building-blocks pattern applied to data products.

The Market Signal

Ivan Zhao reports Notion runs 700 agents alongside 1,000 employees. Jack Clark predicts “great and mostly invisible seas of tokens being used for thinking and exchanging information between silicon minds.” Vasuman is at $3M ARR deploying production agents for enterprise.

The agents are arriving. They need products built for them. Whoever builds the best ones early has a compounding advantage — because agent-consumed knowledge gets better with use, unlike static human-consumed content.

Open Questions