# Jason Lovell

> Frontier AI research, solution design and engineering. Jason Lovell researches what AI can do next, then builds the systems and experiments to find out what holds up. Based in Austin, Texas. Day role: Frontier AI strategy and enterprise innovation, PwC Intelligent Enterprise. Every study and build on this site is his own independent work, built end to end.

Every page has a Markdown twin: add .md to its path (for example https://jasonlovell.ai/work/realitybridge.md), or request it with Accept: text/markdown. The whole site in one file: https://jasonlovell.ai/llms-full.txt

## For agents

- [MCP server](https://jasonlovell.ai/mcp): Streamable HTTP, read-only tools for the profile, studies, builds, releases, search and contact routes. No authentication.
- [Signed tool definitions](https://jasonlovell.ai/tbom.json): the MCP server's Tool Bill of Materials (TBOM v1.0.2), signed with Ed25519. Keys: https://jasonlovell.ai/.well-known/tbom-keys.json
- [How to connect](https://jasonlovell.ai/agents): client configs, and the WebMCP tools the pages register in the browser.
- [Atom feed](https://jasonlovell.ai/feed.xml): studies and releases, newest first.

## Pages

- [Home](https://jasonlovell.ai/index.md): studies, builds and releases at a glance.
- [About](https://jasonlovell.ai/about.md): career, how he builds, what he cares about now.
- [Builds](https://jasonlovell.ai/builds.md): public repositories by area.
- [Releases](https://jasonlovell.ai/releases.md): DOIs and code releases.
- [The studies, as music](https://jasonlovell.ai/listen.md): each study as a track built from its own results, with what every sound stands for.

## Studies

- [MCP Continuation Replay](https://jasonlovell.ai/work/mcp-continuation-replay.md): The action succeeded. The reply vanished. What happens next? recovers, no duplicate.
- [Scaffold Arena](https://jasonlovell.ai/work/scaffold-arena.md): How much of an agent’s performance comes from everything around the model? instrument built.
- [RealityBridge](https://jasonlovell.ai/work/realitybridge.md): A simulator says it worked. Would the real service agree? 3 of 16 diverged.
- [ShadowSkillBench](https://jasonlovell.ai/work/shadowskillbench.md): When an agent learns a procedure, what else does it carry forward? methods preview.
- [Living Context](https://jasonlovell.ai/work/living-context.md): What should an agent revisit when the world changes? refresh plans; learned extension inconclusive.
- [Jev Decision Lab](https://jasonlovell.ai/work/jev-decision-lab.md): What changes when a model returns a typed judgment? workbench.
- [GhostTrace](https://jasonlovell.ai/work/ghosttrace.md): Does a behavioural signal survive repeated self-distillation? toy tier only.
- [No Model Floor](https://jasonlovell.ai/work/no-model-floor.md): Does this benchmark even need a model? headroom 0.
- [CascadeShift](https://jasonlovell.ai/work/cascadeshift.md): Are we measuring the model, or the configuration? interval includes 0.
- [Nano interpretability](https://jasonlovell.ai/work/nano-interpretability.md): When is expensive interpretability worth its cost? wins only when distributed.

## Design records

- [OverhearOps](https://jasonlovell.ai/projects/overhearops.md): What should an agent be allowed to do with what it overhears?
- [TerraRisk Agent](https://jasonlovell.ai/projects/terrisk-agent.md): What would an underwriter need before trusting an agent's risk report?
- [Agent-Ready Checkout Gateway](https://jasonlovell.ai/projects/agent-ready-checkout-gateway.md): What does a merchant need before an agent can buy?

## Contact

- hello@jasonlovell.ai, or the form at https://jasonlovell.ai/#contact. Agents can draft a message there for the person they act for; a person sends it.