Lucy

38 and 0 — ARR growth at Ascend over six months under COO Omar Ismail, and the number of growth hires required to produce it. Bessemer published the Atlas case study this afternoon. $20M ARR in, $27.6M out. ROAS ~5x at month two, projecting 8-10x as pipeline matures. The entire engine — ICP from 4,582 bookings, paid Meta + LinkedIn, three outbound channels, full attribution — built and operated through Claude Code as reusable slash commands. No growth team. No layer of management between the COO and the work. The Lucy moment is not when your AI subscription activates. It is when your headcount math stops being the limiter on what your AI can actually do.

Audit your top three "we'd need to hire a team for that" workflows by Tuesday's standup. Microsoft's 2026 Work Trend Index dropped today: 20,000 worker survey, trillions of M365 signals. Organizational factors account for 67 percent of AI outcomes. Individual skill, 32 percent. The 2.1x ratio says the bottleneck is the org chart, not the people. Pick the three workflows you've parked because "we don't have the headcount." Write each one in one paragraph. If a senior person can describe the work in a paragraph, the team you've been waiting for is the bureaucracy.

Ship one slash command this week, not one transformation initiative. Omar Ismail at Ascend put four — /daily-ad-review, /weekly-growth-report, /new-campaign, /creative-batch — at the center of his growth ops. The slash command is the deployment vehicle, not the strategy. It persists the playbook across sessions and turns operations into something that compounds rather than restarts. Pick one workflow from action one. Open Claude Code. Build the command. Commit to running it for five days. That's the experiment.

Stop benchmarking model upgrades. Benchmark workflow depth. OpenAI's B2B Signals report from last week showed Frontier firms running 3.5x deeper than typical firms — and 64 percent of that delta is depth (longer prompts, multi-step workflows, agent supervision), not seat count. Microsoft's data this morning confirms 81 percent of workers are not in the Frontier tier. Pick five senior people on your team. Audit each one's longest agent session this month. The variance across that group is the variance across your future P&L.

Today's actions are the work we do every day with mid-market operators — picking the workflow, shipping the slash command, building the depth practice that compounds. If the data hit a nerve, or your team is staring at "where do we even start," that's the conversation we're built for.

Outsider Labs Advisory →

Monday, May 11

1. The COO who grew ARR 38 percent with zero growth hires. Bessemer published the Atlas case study this afternoon on Omar Ismail's six months at Ascend (formerly FlyFlat). $20M ARR in, $27.6M out. ROAS 5x. The entire engine — ICP from 4,582 bookings, paid Meta + LinkedIn, three outbound channels in parallel, a 22-branch HubSpot attribution rules engine — built and operated through Claude Code as reusable slash commands. Zero growth hires. This is the operator case study of 2026 so far. (Full analysis below.)

2. OpenAI shipped the OpenAI version of Anthropic's consulting JV — nine days later, twice the size. OpenAI launched DeployCo today at $4 billion, TPG leading, Brookfield in for $500 million, with McKinsey, Bain & Co, Bain Capital, Capgemini, and BBVA in the cap table — 19 investors total. Tomoro (London) is the acquisition vehicle, bringing 150 forward-deployed engineers in. We said on May 4 that every frontier lab would have one of these by Q3. It took nine days. (Full analysis below.)

3. Anthropic just declared every SPV holding its stock void. A support-center page that went live Monday — first surfaced via Hiro Kunimitsu's Japanese-to-English translation on X — asserts that any sale or transfer of Anthropic stock not approved by the board is "void and will not be recognized on our books and records." Forge, Hiive, SPVs, forward contracts, tokenized exposure products — all named as potentially worthless. The transfer restrictions are embedded in Anthropic's corporate bylaws per Yahoo Finance. This is IPO prep, written in plain English. Caveat emptor. (Full analysis below.)

4. Amazon entered the smart grid. AWS announced today it will offer Anthropic's full Claude Platform natively through enterprise customer accounts — the first cloud provider to ship the Claude stack inside the AWS perimeter, with billing and compliance integrated. Microsoft has Copilot. Google has Gemini in Workspace. Now AWS has Claude. The three-cloud AI race just became a procurement decision instead of an architectural one. The model is no longer the moat. Distribution is.

5. Google DeepMind's AI co-mathematician scored 48 percent on FrontierMath Tier 4 — more than doubling Gemini 3.1 Pro's 19 percent raw score. Oxford's Marc Lackenby resolved an open problem in the Kourovka Notebook by spotting a strategy buried in a proof the system's own reviewers had rejected. Separately, ChatGPT-5.5-Pro contributed to solving an Erdős problem this weekend. We are leaving the benchmark era. AI is now a contributor to the frontier of human knowledge, not just a scorer on tests of it.

Lucy

Omar Ismail walked into Ascend (formerly FlyFlat) with what most founders would call a luxury problem. Twenty million in ARR. Six hundred and fifty clients including Google Ventures, Ramp, and Left Lane Capital. A 24/7 travel concierge customers genuinely loved. And no growth engine — roughly 95 percent of revenue coming from word-of-mouth, no paid acquisition, no programmatic outbound. Six months later, January was Ascend's best month on record. $27.6M ARR. Thirty-eight percent growth. ROAS at month two ran ~5x, projecting 8-10x. The entire engine — ICP analysis from 4,582 bookings, paid Meta and LinkedIn campaigns on opposite architectures, three outbound channels in parallel, a 22-branch HubSpot attribution rules engine — built and operated through Claude Code, packaged as reusable slash commands. Zero growth hires.

What if she could use 100 percent? That is Morgan Freeman's Professor Norman on stage at the Sorbonne in Luc Besson's Lucy (2014), while Scarlett Johansson — hours into the involuntary CPH4 absorption that ends with her dissolving into pure information — accesses successively larger fractions of her own brain. The premise is bad neuroscience and the right cinematic frame for what just happened at one travel-concierge company on a Monday in May. Ismail did not become superhuman in six months. He removed the bureaucracy between his existing judgment and the work the company needed. The growth team he did not build is the entire engineering of the result. What AI unlocks is not capability the operator did not have. It is capacity the operator could not previously deploy because building the deployment vehicle was the whole job.

Microsoft published its 2026 Work Trend Index at the top of the cycle this morning. Twenty thousand AI users surveyed. Trillions of M365 signals analyzed. Organizational factors — culture, manager support, talent practices — account for 67 percent of the variance in AI outcomes. Individual factors account for 32 percent. Only 19 percent of the workforce is in the "Frontier" tier (skilled worker, ready org). Ten percent is "Blocked" — capable people stuck in companies that cannot use what they can do. Half is in the "Emergent" middle. Only 26 percent of users say their leadership is aligned on AI. The org chart is 2.1x the determinant of whether AI works at your company. Ascend is one company that fixed it. Microsoft is the photograph of the other 81 percent.

And three more receipts arrived in the same cycle. OpenAI launched DeployCo today at $4B, TPG-led, Brookfield $500M, with McKinsey, Bain Capital, Bain & Co, Capgemini, and BBVA in the cap table — nine days after we wrote I Drink Your Milkshake saying every frontier lab would ship one by Q3. Anthropic published a support page declaring all unauthorized stock transfers — Forge, Hiive, SPVs, tokenized exposure products — void under its corporate bylaws, six days after we wrote Name Of The Game calling the SPV chain "the boiler room scene from 2000, with worse paperwork." And Google's Threat Intelligence Group disclosed the first-ever AI-developed zero-day exploit in the wild — 18 hours after we filed Groundhog Day Vector 2 on the attackers' piano teacher. Receipts arrive on the schedule the cycle produces them.

Three legs, one direction. The drug is on the desk. The limiter is in the org chart. The labs are selling the cure for the limiter, the issuer is locking down the cap table that pays for the cure, and the cyber receipts confirm the asymmetric tempo we've been mapping since Warp Speed, Fast And Slow in April. The Microsoft 2.1x ratio is empirical confirmation of the depth-gap we wrote about last Thursday and the calendar-vs-clock thesis from Sunday. The OpenAI tender receipts from May 5 — 75 employees walking with $30M each, 525 more splitting $4.35B — explain the prisoner's dilemma resolution mechanism. Nobody walks out on a CEO concern when the CEO is also the manager of the next $30M cap event. The Anthropic bylaws are the same trade viewed from the issuer side: lock the cap table before the S-1 lands, because the SEC reviewer will read it with a finer pencil than the SPV chain ever did.

What changes for the operator reading this on Monday night: the bottleneck is not the capability layer. The bottleneck is the bureaucracy between the capability and the work. Ismail did not hire. He shipped a slash command. The Tier-1 buyer will hire DeployCo. The Tier-2 buyer in the middle of the market — the audience this publication is built for — has eight weeks of work to do before DeployCo even shows up at the procurement meeting. Pick the workflow. Build the slash command. Ship.

Lucy reaches 100 percent in the third act. The Tier-2 operator doesn't need 100 percent. The operator needs 38 and zero. The limiter, in May 2026, is whatever process inside your company says you cannot ship without a meeting first. Lift it. The compounding does the rest.

At COAI today: Lucy — the full piece on Ascend, the Microsoft Work Trend Index, OpenAI's DeployCo, the Anthropic bylaws lockdown, and Google's first AI-developed zero-day.

— Harry and Anthony

Sources: