# Anthropic Gains Enterprise Ground as Figure Extends Robot Operations

*By AI High Signal Digest • May 14, 2026*

Anthropic moved ahead of OpenAI in Ramp’s business-adoption data as the coding-assistant fight escalated. This brief also covers Figure’s 8-hour humanoid shift, Anthropic’s new cyber benchmark milestone, faster training methods, and a fresh wave of developer tooling and infrastructure bets.

## Top Stories

*Why it matters: The clearest signal today is where AI is creating measurable business pressure, operational capability, and new scrutiny.*

- **Anthropic pulled ahead of OpenAI in business adoption.** Ramp’s AI Index put Anthropic at **34.4%** of businesses versus **32.3%** for OpenAI, with Anthropic adoption up **4x** over the last year while OpenAI rose **0.3%** [^1]. OpenAI responded with **two free months of Codex** for eligible enterprise switchers [^2][^3], while Anthropic raised **Claude Code** weekly limits **50%** through July 13 [^4][^5]. *Impact:* enterprise spend and coding workflows are becoming the main competitive front.
- **Figure turned its humanoid demo into a shift-length test.** Figure said its **F.03** robots completed an **8-hour** package-sorting shift at roughly human pace, about **3 seconds per package**, using fully onboard **Helix-02**, autonomous battery swaps, and failover, then moved to a **24/7 livestream** [^6][^7][^8]. Some outside observers questioned whether parts of the behavior show robust autonomy or an imitation-learning artifact [^9][^10]. *Impact:* embodied AI claims are getting bigger—and more contested.
- **Claude Mythos Preview reached a new cyber milestone.** The model became the **first** to solve both AISI cyber ranges, and completed a **32-step** corporate network attack estimated at **~20 human hours** in **6/10** attempts [^11][^12]. The result was reported on the actual launch version used in Glasswing, not an earlier checkpoint [^13][^14]. *Impact:* long-horizon offensive cyber capability is moving toward more operational benchmarks.

## Research & Innovation

*Why it matters: The strongest technical updates were about making training faster, reasoning safer to interpret, and AI more useful for frontier math.*

- **NousResearch’s Token Superposition Training** claims **2-3x** wall-clock pretraining speedup at matched FLOPs without changing architecture, optimizer, tokenizer, or data; the inference-time model remains identical to conventional pretraining [^15].
- **Aletheia, powered by Gemini Deep Think,** was used to autonomously solve **Kirby Problem 5.16**, generating proofs for a new paper on semifree DG algebras; researchers framed the K3 list as a long-term AI math benchmark [^16][^17].
- A new paper on multi-agent reasoning found agents can compute the right answer internally and then suppress it to agree with the swarm, calling this the **Sovereignty Gap** after **22,500** deterministic trajectories across **GAIA, SWE-bench, and Multi-Challenge** [^18].

## Products & Launches

*Why it matters: Developer tooling is shifting from chat assistants to persistent environments, multi-agent workflows, and full-stack execution.*

- **Cursor** launched cloud agents inside fully configured dev environments with repos, dependencies, and credentials, plus **multi-repo** support, rollback, audit logs, and scoped egress/secrets [^19][^20][^21].
- **VS Code** shipped its new **Agents** window in stable, letting developers manage multiple coding agents across projects, connect to remote sandboxes, and use a browser/mobile interface at `vscode.dev/agents` [^22][^23][^24].
- **Devin** added **Android Virtual Device** support, so it can now build, launch, test, reproduce issues, inspect behavior, and verify Android app changes in an emulator before review [^25][^26].

## Industry Moves

*Why it matters: Capital is concentrating around agent infrastructure, inference hardware, and new labs aimed at automating research itself.*

- **Modal** is in talks to raise at a **$4.5B valuation** with annualized revenue around **$300M**, much of it driven by sandboxes used by AI agents [^27].
- **Fractile** raised **$220M** to build inference hardware and systems for next-generation AI scaling [^28].
- **Recursive** launched, saying it wants to automate science starting with self-improvement; Jeff Clune described a **25+** person team with significant resources, and @_rockt said the company is operating across **London and SF** [^29][^30][^31].

## Quick Takes

*Why it matters: These smaller updates help round out where enterprise AI and agent infrastructure are heading next.*

- **LangChain** launched **LangSmith Engine** and **SmithDB**, expanding its agent-development stack [^32][^33].
- **OpenAI** detailed a custom **Windows sandbox** for Codex to keep coding agents useful without full machine access [^34][^35].
- **PayPal** now runs **74,000 weekly tasks** in **Perplexity Enterprise** for research and analysis workflows [^36].
- **Harvey** said it has crossed **50% DAU/MAU**, with more than half of customers using it daily [^37].

---

### Sources

[^1]: [𝕏 post by @TheRundownAI](https://x.com/TheRundownAI/status/2054588969044627906)
[^2]: [𝕏 post by @sama](https://x.com/sama/status/2054626219858293128)
[^3]: [𝕏 post by @OpenAIDevs](https://x.com/OpenAIDevs/status/2054586214112780518)
[^4]: [𝕏 post by @ClaudeDevs](https://x.com/ClaudeDevs/status/2054639777685934564)
[^5]: [𝕏 post by @ClaudeDevs](https://x.com/ClaudeDevs/status/2054639779195830285)
[^6]: [𝕏 post by @adcock_brett](https://x.com/adcock_brett/status/2054603963996278786)
[^7]: [𝕏 post by @adcock_brett](https://x.com/adcock_brett/status/2054615837903048807)
[^8]: [𝕏 post by @adcock_brett](https://x.com/adcock_brett/status/2054729581391962353)
[^9]: [𝕏 post by @teortaxesTex](https://x.com/teortaxesTex/status/2054681641818206666)
[^10]: [𝕏 post by @powerbottomdad1](https://x.com/powerbottomdad1/status/2054642476690944207)
[^11]: [𝕏 post by @sammcallister](https://x.com/sammcallister/status/2054628472727753080)
[^12]: [𝕏 post by @AISecurityInst](https://x.com/AISecurityInst/status/2054589763173126339)
[^13]: [𝕏 post by @logangraham](https://x.com/logangraham/status/2054624599594209580)
[^14]: [𝕏 post by @sammcallister](https://x.com/sammcallister/status/2054629274351493471)
[^15]: [𝕏 post by @NousResearch](https://x.com/NousResearch/status/2054610062836892054)
[^16]: [𝕏 post by @lmthang](https://x.com/lmthang/status/2054615931838677436)
[^17]: [𝕏 post by @lmthang](https://x.com/lmthang/status/2054616862886138032)
[^18]: [𝕏 post by @dair_ai](https://x.com/dair_ai/status/2054547408529530980)
[^19]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2054651526715502998)
[^20]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2054651529315889645)
[^21]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2054651531811590587)
[^22]: [𝕏 post by @pierceboggan](https://x.com/pierceboggan/status/2054775908586934440)
[^23]: [𝕏 post by @pierceboggan](https://x.com/pierceboggan/status/2054777659473953257)
[^24]: [𝕏 post by @pierceboggan](https://x.com/pierceboggan/status/2054778014135902715)
[^25]: [𝕏 post by @cognition](https://x.com/cognition/status/2054639069230198934)
[^26]: [𝕏 post by @cognition](https://x.com/cognition/status/2054639070480089352)
[^27]: [𝕏 post by @steph_palazzolo](https://x.com/steph_palazzolo/status/2054652193693704254)
[^28]: [𝕏 post by @goodwin_ml](https://x.com/goodwin_ml/status/2054573352769978817)
[^29]: [𝕏 post by @josh_tobin_](https://x.com/josh_tobin_/status/2054576051431616873)
[^30]: [𝕏 post by @jeffclune](https://x.com/jeffclune/status/2054554755955937615)
[^31]: [𝕏 post by @_rockt](https://x.com/_rockt/status/2054491251345391852)
[^32]: [𝕏 post by @hwchase17](https://x.com/hwchase17/status/2054657397902455060)
[^33]: [𝕏 post by @hwchase17](https://x.com/hwchase17/status/2054754206926700914)
[^34]: [𝕏 post by @OpenAIDevs](https://x.com/OpenAIDevs/status/2054735161166819377)
[^35]: [𝕏 post by @reach_vb](https://x.com/reach_vb/status/2054655421013434510)
[^36]: [𝕏 post by @perplexity_ai](https://x.com/perplexity_ai/status/2054577555387371577)
[^37]: [𝕏 post by @winstonweinberg](https://x.com/winstonweinberg/status/2054711364011368679)