Spine AI

The harness is the capability multiplier. Models are locally capable. They can reason, code, search, and use tools on the step in front of them. But ambitious outcomes are global: they depend on the plan, the handoffs, the context that survives them, which model handles each task, and what gets verified before the system is allowed to stop. We built Medley for that layer. The 'mission' harness has three core components: 1. a planner agent; 2. a contract-driven DAG of workers, where each node can use a different model/harness pair and its own instructions; and 3. a review agent that tests the assembled result against the contract and triggers focused repair work when it falls short. A separate benchmark meta-harness learns the benchmark-specific instructions. It selects unlabeled tasks, runs the mission harness, studies its traces, and edits the responsible planning, worker, review, or runtime skill. The adaptation is unsupervised and grade-blind: it receives no solved examples, labels, or grader feedback. The reason for the graph is simple: you cannot fully plan long-horizon work in advance. The information needed to solve it emerges during execution. You need an operating system that can update the graph without losing the contract. The resulting systems set new reported highs across terminal work, clinician conversations, full-stack applications, and drug discovery. Full report: medley.sh/blog/the-harness-i…

Today, we’re launching /mission by Medley. Give Claude Code an outcome, not a task. Medley turns it into a live graph, coordinates Claude Code + Codex workers, then reviews and keeps going. SOTA on 4 public benchmarks—from coding to healthcare and drug discovery. medley.sh/

This week as a CEO. I was the head of sales, the recruiter, the growth lead, and the customer success team. People think the hard part of founding is wearing every hat. It’s not the wearing. It’s that each hat is a full-time job; and there is exactly one of me.

everyone's moving from markdown to html/mdx. karpathy talked about having LLMs generate images + slideshows. thariq wrote this playbook for it. let me save everyone 2-3 months of discovery: having a harness where LLMs output typed, human-auditable artifacts works. it's net better than a traditional agent harness. we compared opus 4.7 using claude's vanilla agent sdk vs opus 4.7 using spine's harness with typed artifact blocks. same model. different harness. spine won 89% of decisive pairs across 140 deep research evals. the reason this works is not just "better outputs." typed artifacts make long-running tasks way easier for the harness to manage. context passing across block types gets standardized. big artifact type nobody's called out yet: tables / excel / lists as structured intermediaries. once you have structured state between steps, downstream blocks can do sql-like operations instead of re-reading another wall of prose. and the bonus: humans get useful, editable, auditable intermediate artifacts along the way. not just a final answer. if your LLMs aren't generating intermediary artifacts yet, you're leaving a lot on the table.

This is just the beginning. Spine's coming up on 3 years. Back then you couldn't do any of this. Now you can imagine what you want and make it happen. I know so many people (myself included) who have swapped video games for working with agents. What a time to be alive. Biggest thing I've learned building through this: planning for the status quo isn't enough. You need to think about where your product, your customers, and the market will be 4-5 months from now. That extra dimension of thought didn't matter before. It does now. Customers aren't buying the current version of your product. They're buying into your company's ability to keep up with progress. I don't think enough founders are internalizing this.

Night and day talking to folks in SF vs. everywhere else. SF is whiplashing between OpenClaw, Nous Research, open-sourced Claude Code clones. New drops on X every week, everyone scrambling to figure out how it fits. This week I talked to customers in Denver, Toronto, Boston, Dubai, Mumbai. They've all used ChatGPT. Most just started hearing about Anthropic. No idea what OpenClaw is. They don't care. What they care about is seeing what the AI is doing, tracing what happened, and getting better outcomes on the first try. Most grounding conversations I have all week. SF is debating infrastructure wars while the rest of the world is just getting real work done. Vibe coding is the tip of the iceberg. The bigger pie is non-technical knowledge work. And given that building is commoditized, the rest of the stack matters even more.

The most productive hour of my day is 3:00 AM. I am fast asleep. Just launched. No research team. No analyst. No budget for one. I do have Spine AI Automations. Every morning I wake up to fresh intelligence reports on my space. Prompt. Set interval. Schedule. That's it. Spine AI runs the research. I run the company.