# Brian Sparker

> Brian Sparker’s personal website. Explore my work in AI, search, and developer tools, browse my projects and writing, or ask my AI about my experience.

Staff Product Manager at You.com. Previously Crate & Barrel and G2. MBA, Boston University.
This is Brian’s personal website, replacing sparker.co. Brian is not offering consulting services. Older writing about consulting is historical.

## Explore
- [About Brian](https://sparker.ai/about): Background and interests.
- [Work](https://sparker.ai/work): Product work at You.com, Crate & Barrel, and G2.
- [Writing](https://sparker.ai/field-notes): Notes on AI, search, and product thinking.
- [Talk to Brian’s AI](https://sparker.ai/ask): Answers grounded in his published work, with sources.
- [Contact](https://sparker.ai/contact): Email Brian directly at brian@sparker.ai.

## For agents
- MCP server (Streamable HTTP, no auth, read-only): https://sparker.ai/mcp. Tools: ask_brian, search_site, list_writing, read_page, get_profile.
- Ask endpoint: GET https://sparker.ai/api/ask?q=... or POST {"question": "..."} returns {"answer", "sources"} as JSON.
- Every piece below is linked as markdown; drop the .md suffix for the HTML page.
- Setup instructions: https://sparker.ai/agents

## Side projects
- [SuperPaste](https://github.com/brainsparker/superpaste): Press Option+V and the right text appears in whatever you are typing. Native macOS, Swift, MIT.
- [frugal](https://github.com/brainsparker/frugal): The open routing layer for AI tool calls: every search, extract, and browse call routed to the cheapest provider that works, with failover and a receipt. One Go binary.
- [you.md](https://github.com/brainsparker/you.md): One portable preferences file for AI coding tools instead of six that disagree. Open protocol, TypeScript, MIT.
- [PromptLens](https://github.com/brainsparker/PromptLens): Prompt and agent evaluation from one YAML golden set across Claude, GPT, Gemini, and local models, with cost and latency in one report. Python.
- [LabelMerge](https://github.com/brainsparker/free-mail-merge): CSV in, mail-merged PDF out, entirely in the browser. Your data never leaves your machine.
- [GTM skills](https://github.com/brainsparker/skills): Claude skills for positioning, messaging, and narrative work, so an AI assistant applies the frameworks consistently.

## Writing, frameworks, and bets
- [Agents search like it's 1999](https://sparker.ai/field-notes/agents-search-like-its-1999.md): Coding agents now pick tools while a developer watches, and the queries they send look nothing like a buyer's. They are terse, literal, and dated. If your pages only answer the human query, the agent never finds you.
- [Attribution is who called, never what they asked](https://sparker.ai/field-notes/attribution-is-who-called.md): Most agent traffic reaches an API through SDKs, plugins, and MCP servers that never identify themselves, so the channels driving adoption look like they do nothing. The fix is a client identifier everywhere, with one line you do not cross.
- [The router is the moat, the rail is a feature](https://sparker.ai/field-notes/the-router-is-the-moat.md): Agent payments will not settle on one protocol soon. Instead of betting on a winner, build the router: one price, one ledger, and many rails underneath that you can promote or demote without touching anything else.
- [Phone Computer: will workspace habits become consumer habits?](https://sparker.ai/bets/phonecomputer-ai.md): A bet that workspace habits will become consumer habits: people with different personas will use task-specific assistants to get the same jobs done.
- [The next bet is Phone Computer](https://sparker.ai/newsletter/the-next-bet-is-phone-computer.md): The consulting bet will close as failed on its original deadline. The next bet is that workspace assistant habits will become consumer habits.
- [The Live Context Layer for AI](https://sparker.ai/field-notes/the-live-context-layer-for-ai.md): Models are becoming interchangeable. Knowing what is true right now is not. The next major AI infrastructure company may look less like another model lab and more like Bloomberg for the open web.
- [Permissionless aggregation: what x402 changes](https://sparker.ai/field-notes/permissionless-aggregation-x402.md): AI gateways already consolidate credentials, budgets, policy, spend, and routing. x402 matters because it can make new providers purchasable without a bilateral account or catalog deal.
- [Define the decision before you build the AI pilot](https://sparker.ai/field-notes/decision-criteria-before-ai-pilots.md): A pilot without a decision rule can demonstrate technical possibility while leaving the organization no closer to knowing what deserves investment.
- [Can Sparker.ai sell an AI Product Decision Sprint?](https://sparker.ai/bets/sell-ai-product-decision-sprint.md): Failed-bet assessment recorded; scheduled to close September 30. Referrals and slow problem recognition made the consulting sales cycle a poor fit for the pace of AI.
- [The consultancy is the first Bet](https://sparker.ai/newsletter/the-consultancy-is-the-first-bet.md): Issue 001 introduces Sparker.ai’s operating model and puts its first commercial assumption under the same evidence standard it recommends to clients.
- [The Platform and Bets Decision Framework](https://sparker.ai/platform/platform-and-bets-decision-framework.md): A lightweight operating system for protecting dependable work while giving uncertain ideas a disciplined path to earn investment.
- [Decision Principles](https://sparker.ai/principles/decision-principles-v0-1.md): The canonical principles behind how Sparker.ai evaluates uncertainty, runs experiments, ends work, and turns decisions into reusable assets.