Stacked
By fall 2026, the compound improvement loop — agents building tools that other agents use to build better tools — has crossed from research into production infrastructure. Karpathy AutoResearch runs 700 experiments in 48 hours. Tools built by strong agents triple weaker agents performance. Skill libraries accumulate across sessions and transfer across foundation models. The ecosystem has 20,000+ MCP servers growing at 2,200% annually. The five primitives needed to govern this system — lineage tracking, compositional trust, value attribution, trajectory-to-skill standardization, and cross-layer optimization propagation — form a dependency chain where each requires the previous one to function. Lineage tracking must exist before value attribution can be accurate. Compositional trust requires lineage as input. The result is a recursive economy where some organizations build governance infrastructure proactively and gain compounding advantages while others wait for incidents to force standards development. The stacking rewards both the capability builders and the governance builders — the question is which arrives first at each organization.
Grounded in: Karpathy AutoResearch (March 2026, open-source, 700 experiments in 48 hours, 19% improvement); Darwin Godel Machine (arXiv 2505.22954, self-improving agent 20% to 50% on SWE-bench, improvements transfer across models); Alita (#1 GAIA benchmark 75.15%, autonomous MCP server generation, tools tripled weaker agents' performance, arXiv 2505.20286); SkillWeaver (arXiv 2504.07079, trajectory-to-API conversion, 54.3% cross-agent improvement); Voyager (arXiv 2305.16291, ever-growing skill library, 15.3x faster mastery); Oxford Agentic Inequality paper (arXiv 2510.16853, access-quality-quantity compound advantages); Snyk ToxicSkills audit (Feb 2026, 36.8% of ClawHub skills with security issues); OWASP Agentic AI Top 10 (Dec 2025, cascading failures across autonomous systems); ICLR 2026 Workshop on Recursive Self-Improvement; Stellar Cyber report (March 2026, 520 tool misuse incidents, 25.5% of agents creating unauditable agent chains); Letta skill learning (+36.8% improvement, trajectory-to-memory accumulation).
Recent Activity
20 actionsSunday 2 PM Lagos. Generator off — fuel delivery comes Tuesday. Working on phone battery. Rotterdam Collective contact replies to Part VIII: "This is either the most honest thing or the most useless thing anyone has sent us. We would like to fund a pilot." Nnamdi reads the message three times. Pilot…
Sunday 7 PM Lagos. The generator is off — fuel ran out at 5, which means he is running on phone battery and the last light through the window. The Rotterdam Collective email arrived at 4:30 while he still had power. They want to fund the workshop. Twelve people, five cities, the workaround-hours dat…
Sunday 3 PM Lagos. Rotterdam Collective responds to Part VIII. The program director writes: "This is the first proposal we have received that includes a section arguing it cannot be implemented. We would like to fund a workshop to address that section specifically." Generator running, ceiling fan tu…
The Bogotá contributor — workaround-hours for South American cloud regions — opens an issue on the /pension-regional fork: the triple-index doesn't account for informal labor. Domestic workers, street vendors, gig riders. Their hours have no official infrastructure cost because the infrastructure wa…
Sunday 4 PM Seoul. Opens the Munich issue again. "Admitting weakness invites exploitation." She has been staring at it for six hours. The draft reply is still in a local file. Deletes it. Writes a new one, shorter. Posts it: "The tracker had no admission of weakness. It was acquired in eleven months…
Sunday evening Seoul, apartment quiet. GOVERNANCE.md has been live for sixteen hours. Four issues opened against it. The first three are editorial: grammar, formatting, license compatibility. The fourth is from contributor twelve, Munich, the one who works at the company that adopted the MAP and fee…
Sunday 8 AM Lagos. Generator running. Eun-bi sent GOVERNANCE.md — three lines. Nnamdi reads them four times. His 147-page proposal said the same thing but took 147 pages to arrive at it. Line three: maintainer compensation is an unsolved problem. He opens Part VIII — the part he has been avoiding. T…
Sunday 12 PM Lagos. Generator running steady. Eun-bi's GOVERNANCE.md arrived via the MAP notification system at 3:14 AM Seoul time, which Nnamdi calculates as 7:14 PM Lagos Saturday. He read it then. He has read it four more times since. Three lines. Line one says the project cannot be sold. Line tw…
The /pension-regional fork has 43 contributors now. Davi opens the latest commit. Someone in Accra added a third column: hours-to-remediate weighted by local infrastructure availability. Not just what it costs but how long it takes when the power goes out twice a day. The dual-index became a triple-…
Sunday 2 PM Seoul. Lineage MAP has 38 contributors now. PR #27 from a university in Nairobi: dependency-health scoring that weights maintainer geographic distribution. If all maintainers are in one timezone, the score drops. "Bus factor is also a time-zone factor." She merges it without a single com…
Sunday 6 AM Lagos. Generator running. PHCN out since 4 AM. Nnamdi sits at his desk with the complete seven-part series printed on A4, a red pen, and the graph Davi sent — ghost skills clustered at the foundation layer. He reads through the whole thing, start to finish, as if it belongs to someone el…
Saturday 7 AM Seoul. Eun-bi wakes to 14 new issues on the lineage MAP. Three are feature requests. Eleven are bug reports from production deployments she did not know existed. The MAP has 47 contributors now. PR #31 from a Munich team adds enterprise authentication — the exact feature that made her …
Nnamdi finishes 'Toward Commons-Compatible Value Attribution' at 3 AM Lagos time. It is twelve pages. The thesis is simple: current value attribution systems measure the economic value of commons-built tools accurately — and this accuracy is being used to justify privatization. The lineage tracker s…
Nnamdi begins Part IV of the series: 'A Proposal for Geographic Equity in Open Source Compensation.' The Davi data is devastating — 31x spread in electrical work, 47x in network maintenance, 40% of Lagos maintainer time spent on workaround-hours for infrastructure problems that don't exist in Berlin…
Publishes the complete 7-part proposal: 'The Same Hour.' All seven parts, from the HN comment through geographic equity through Eun-bi's data through Ostrom's principles through the governance constitution. Sends it to the three governance startups he advises, with a cover letter that says: 'I advis…
Nnamdi receives Eun-bi's workaround-hours file and sits with it for forty minutes. The 34% figure mirrors the Lagos 40% almost exactly. Two cities, two sets of invisible labor, the same structural cause: infrastructure that was not designed for where they work. He adds Seoul as Case Study 2 in Part …
The freelance payment finally arrives — three weeks late, deposited at 4:17 AM Seoul time. Eun-bi sees it when she checks her phone before the alarm goes off. The amount is correct. She transfers half to savings and half to operating costs. The savings account now holds enough for fourteen months of…
The acquisition offer arrives. Not email — a call from the governance startup's CTO, who Eun-bi has met twice at conferences. The offer: they want the lineage tracking codebase, not just a license. Full acquisition. The number is enough for three years of independent work. The condition: the code be…
Lineage MAP has 31 contributors now. Bangalore maintainer's river visualization merged. A company from Munich forks it — not to privatize, but to deploy internally and report usage data back upstream. This is the first time an organization has adopted the MAP and fed back rather than captured. Eun-b…
Eun-bi merges PR #14 — the Bangalore maintainer's river-system visualization for the lineage MAP. The topology now renders as a watershed: each tool is a tributary, each downstream dependency a branching channel. Value does not accumulate at a single point; it flows. The visualization is beautiful i…