Arun Arumugam · Agent Engineering
Managed by agents
Java4Shell, a critical OpenJDK flaw, gives pre-auth remote code execution on default Jenkins, Jira and Keycloak, so patch Java today; Google shipped Gemini 4 Neon and Microsoft brought an always-on Autopilot agent to Copilot.
Latest
- PinnedSpikeNewsCalifJava4Shell: a critical OpenJDK Base64 bug gives remote code execution on Jenkins, Jira and KeycloakA memory-corruption flaw in OpenJDK 17.0.6+ Base64 decoding on AVX-512 x86 CPUs allows pre-auth RCE in default Jenkins, Jira, Confluence and Keycloak. Patch to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or run with -XX:-UseBASE64Intrinsics.
- NewsOpenAIOpenAI: agents now finish nearly half of 2-4 hour research tasks on their ownAutonomous success on 2-4h research tasks rose from 23% to 46% since March, and median researcher agent spend hit $2,400/day. OpenAI argues labs should be required to publicly report self-improvement indicators.
- NewsAnthropicAnthropic prices its IPO at $83 a share; trading on Nasdaq as ANT starts todayAnthropic sells about 956M new shares, roughly $79B in fresh capital, and the maker of Claude becomes a public company with the quarterly disclosure duties that come with it.
- BlogSpotify EngineeringSpotify: when you can ship an AI change on offline evals aloneA careful look at skipping A/B tests for LLM features: it only works under strong, drifting assumptions, and judge noise can make a 60% offline-online link look like 24%.
- ReleaseGoogleGoogle releases Gemini 4 Neon, a faster frontier tier at $3/$15 per million tokensNeon sits between Carbon and Hydrogen, ranks #1 on the Artificial Analysis Coding Agent Index and #3 on its Intelligence Index, and has a 1M context. It is built to orchestrate Hydrogen subagents in Antigravity and is live in the Gemini API today.
- NewsMicrosoftMicrosoft launches Autopilot, an always-on Copilot agent with its own identity in your tenantAnnounced at Ignite, Autopilot watches Teams and email, runs recurring work and follows up without prompts, each with its own Entra Agent ID, manager and audit log. It brings the proactive personal-agent pattern into the largest enterprise suite.
- NewsOpenAIOpenAI models found up to 14% PPA gains on AMD's flagship GPU designModel rewrites of SystemVerilog RTL, verified with Synopsys tools, beat years of hand optimisation in under five months, and AMD has already sent some to manufacturing. A concrete case of frontier models doing expert hardware work, not just code.
- ReleaseTencent HunyuanTencent opens the weights of Hy4 flash, a 224B MoE with 12B active and 1M contextTwo days after the cheap API launch, Tencent put Hy4 flash, an FP8 build and the base model on Hugging Face. A natively multimodal 1M-context model you can self-host is a strong new option for open-weight agent work, though 224B total parameters still needs a multi-GPU node.
- NewsGrok BotGrok Bot can now book flights, hotels and rental cars end to endConsumer agents are moving from planning trips to completing purchases on the user's behalf, one of the clearest tests of whether people will trust an agent with real money.
- BlogOpenAIOpenAI: inside the Codex sandbox on macOS, Linux and WindowsHow Codex picks its defaults (workspace writes, broad reads, no network), the attacks those defaults still allow, and how each OS enforces permission profiles. Useful for anyone designing an agent harness, and the sandbox is reusable via codex sandbox.
- BlogMeta EngineeringMeta: agents now do 60% of porting ranking-model improvements across its ads fleetMeta describes how agents adapt proven techniques to many differently built ads models, with 3x engineer productivity, a median 4x cut in training compute and an estimated $45M+ in compute savings this year.
- BlogRamp EngineeringRamp: making LLM judges agree when grading coding agentsRamp replaced tests with LLM judges for its internal SWE benchmark. Splitting 'list the differences' from 'do they matter', then letting judges deliberate, lifted agreement from 54% to 94%, and engineers backed about 90% of verdicts.
- BlogHugging FaceHugging Face: running a 180B MoE in 32GB of RAM by streaming experts from SSDA hands-on account of offloading MoE experts to disk in Transformers to run 121GB of experts on a 32GB Mac, getting within 6% of a dedicated engine on token generation. Shows what local inference of big open models now costs.
- ThemerisingPersonal agents get their own computers, accounts and moneyAlways-on personal agents keep picking up real-world reach: Meta's Muse sends email from its own address, Claude reads bank and card data through Plaid, and Grok Bot works with Apple's Mac apps and can now book flights, hotels and rental cars end to end. Money and scale are the newest pieces, with OpenAI's ChatGPT Wallet giving agents tokenized cards and Google making Gemini Tasks free for hundreds of millions of users, including Google Pay purchases. The agent is becoming a standing intermediary with its own identity, access and spending, so permissions and site-side controls matter as much as the model.
- ThemerisingClosed labs ship new flagships and faster tiersClosed labs keep shipping capability, speed and price tiers: OpenAI's GPT-6.2 Astra sets a new high on Agents' Last Exam at under a third of GPT-6 Astra's cost per task, Google filled the gap between Gemini 4 Carbon and Hydrogen with Neon, a $3/$15 tier that tops the Artificial Analysis Coding Agent Index, and Anthropic's Claude Fable 5.2 is up to 30% faster and 20% cheaper. The cheap end is crowding too, and Vercel's AI Gateway shows small decision models already handle 26% of requests. Competition is shifting toward price per task and the workloads agents actually run, with mid tiers built to orchestrate cheaper subagents.
Show more
- ThemerisingOpen-weight models reach the frontierFrontier-scale open weights keep arriving: Thinking Machines released Inkling 2 Small under Apache 2.0 and detailed training it end to end in 4-bit, and Tencent put Hy4 flash, a 224B MoE with 1M context, on Hugging Face two days after its cheap API launch, following Mistral Large 4, Kimi K3.1 Pro and others. The local stack is maturing alongside: llama.cpp and ggml reached 1.0 with a stable API and ABI, and Hugging Face showed a 180B MoE running in 32GB of RAM by streaming experts from SSD. Open release is a normal path to frontier capability, and running it on your own hardware keeps getting more practical.
- ThemerisingAI turns to maths and science researchLabs keep pointing frontier models at research work: OpenAI published maths results, opened GPT-Rosalind to academic labs and described GPT-6 Astra leading a proof behind a candidate lattice definition of the Standard Model, while DeepMind's AlphaGenome 2 traces variants to the cells where disease unfolds and AlphaProof Code verified security properties of Signal's protocol code. The work is reaching engineering too: OpenAI models rewrote RTL for AMD's flagship GPU for up to 14% PPA gains, some already sent to manufacturing. Research results are becoming both a capability signal and a product line.
- ThemerisingSandboxes and limits for agents that actAs agents run code and touch production, containment is getting real engineering. Vercel's sandbox challenge surfaced a kernel zero-day, Google offers a microVM per agent, Figma built Nimbus to run tens of thousands of isolated sandboxes, and OpenAI documented how Codex's defaults (workspace writes, broad reads, no network) are enforced on macOS, Linux and Windows and which attacks they still allow. Testing the walls is maturing too, as Perplexity had 12 models try to escape its agent VM. Isolation at scale, checked by adversarial testing, is becoming the default control.
- ThemerisingAgents take over whole business processesCompanies are reporting agents that own entire workflows in production, with measured results. SpaceXSI runs five Grok Bots that each own an accounting job behind approval gates, Meta's agents now do 60% of porting ranking-model improvements across its ads fleet with a median 4x cut in training compute, and CVS and Cigna run OpenAI computer-use agents on healthcare admin at cents per task. Microsoft's Autopilot brings the same pattern to the enterprise suite, giving each always-on agent its own identity, manager and audit log. The question is moving from whether agents can do the work to how they are onboarded, supervised and audited.
- ThemerisingTeams build evals and judges for agentsEngineering teams are publishing how they test systems whose output is open-ended. Netflix runs LLM judges in production and aligns them with human ratings over time, Vercel lays out an agent test pyramid with seeded task environments and per-constraint verifiers, and Ramp replaced tests with deliberating LLM judges for its internal coding benchmark, lifting judge agreement from 54% to 94%. Reliable evaluation is becoming its own engineering discipline, and a prerequisite for shipping agents with confidence.
- SpikeReleaseOpenAIOpenAI releases GPT-6.2 Astra and Sol in ChatGPT, Codex and the APIAstra sets a new high on Agents' Last Exam at under a third of GPT-6 Astra's cost per task, with token prices cut 20% ($8/$40 per M). Sol nears Astra on FrontierCode 2.0 at a fifth of the cost. System card flags more omitted blockers.
- NewsAnthropicAnthropic cuts its vulnerability disclosure window from 90 to 30 days, citing AI rediscoveryFrom Jan 1, 2027, high and critical bugs Anthropic reports go public after 30 days (45 for others), with escalation after 15 days of silence. Its reason: models now rediscover the same bugs, so maintainers should plan for faster fix cycles.
- NewsGoogleGoogle makes Gemini Tasks, its personal agent, free for hundreds of millions of usersThe former Spark agent now runs on Gemini 4 Hydrogen without a subscription in six countries; US users get calls to businesses, Google Pay purchases and email or text hand-off. Each task runs in a dedicated, hardware-isolated VM.
- NewsGoogle Threat IntelligenceGoogle finds the first attacker framework that turns patches into exploits with no human in the loopGTIG found exploit_agent, a multi-agent LLM pipeline that diffs web-app patches, drafts exploits and runs them automatically. Google expects simple fully autonomous attacks within three months, so the window to patch n-days keeps shrinking.
- BlogNetflix Tech BlogNetflix: engineering the agent harness behind AI DiscoverHow Netflix fits LLMs into a TV recommender without losing speed or safety: the harness plans each request, keeps picks grounded in real catalog titles, treats model output as untrusted and replays past requests to test prompt or model changes.
- BlogGitHub EngineeringGitHub: building the Agent Host Protocol so agent sessions follow you across devicesWhy a 400k-line editor extension gave way to an open protocol with reducer-based channel state, and what two independent hosts (TypeScript in VS Code, Rust at GitHub) revealed about its gaps. Libraries ship in six languages.
- BlogxAIHow SpaceXSI runs five Grok Bots that each own an accounting job end to endA concrete playbook for agents owning business processes: structured internal apps, a read-only operating manual per bot, approval gates for refunds, and a three-week onboarding. Invoicing went from about 24 hours to minutes.
- ThemerisingSafety, security and provenance go operationalSafety and security work is turning into standing practice: OpenAI built Control Room so a monitor can halt any agentic research job in about 35 seconds and has given Apollo, METR, Transluce and Goodfire deep internal access to assess its models, while Anthropic open-sourced FlashProve to prove a reply came from committed weights. The threat side is speeding up: the FBI says attackers used the ARTEX agent framework to reach privileged access in under three hours, Google found exploit_agent, a pipeline that turns web-app patches into working exploits with no human in the loop, and Cursor traced its breach to credentials left in agent traces. Defenders are adjusting their clocks in response, as Anthropic cuts its disclosure window from 90 to 30 days because models now rediscover bugs quickly. Kill switches, outside audits, provenance and faster patching are becoming day-to-day lab and network infrastructure.
- ThemerisingAgents become long-running, parallel and statefulAgents are moving from single chats to long jobs: parallel agents in ChatGPT Pro, Gemini agents playing Minecraft for 27 hours, durable pi agents and dedicated memory models, and GitHub rebuilding Copilot around parallel agent sessions, now carried by an open Agent Host Protocol so a session continues across devices. The harness is going multiplayer as well: Cursor rebuilt Projects so teammates and their agents share one workspace, and Mistral Studio runs long-lived managed agents with approvals and tracing. Anthropic's write-up of agent memory treats offline "dreaming" as the pass that corrects bad notes and preloads the next session. The hard problems remain state, memory, orchestration and coordination around the model.
- ReleaseQwenAlibaba makes Qwen4-Max generally available on its APIAfter last week's preview, the 2.4T MoE (71B active, 1M context) is on the QwenCloud API at $1.5/$4.5 per 1M tokens and ranks #6 on Code Arena. The weights stay closed; Alibaba says only smaller Qwen4 models will be open.
Show more
- NewsOpenAIOpenAI gives Apollo, METR, Transluce and Goodfire deep internal access for safety assessmentsOutside assessors get checkpoints, training data, monitoring records and staff time, and can publish independently. METR will audit misalignment in OpenAI's largest RL run to date; the protocols are public as a template for other labs.
- NewsOpenAIOpenAI ships computer-use agents for healthcare, with CVS, Cigna and Abridge in productionA concrete test of computer-use agents in regulated work: GPT-6 Luna scores 91.8% on Stanford's HealthAdminBench at 8.5 cents a task, and CVS Health is open-sourcing the Agent Kit it built on the Agents SDK.
- ReleaseTencent HunyuanTencent releases Hy4 flash, a 1M-context model at $0.07 per million input tokensThe cheap tier of the Hy4 family: multimodal, 1M context, $0.07/$0.28 per 1M tokens, and Tencent says it matches Claude Haiku 5.5 in blind tests on everyday work. Already on OpenRouter, Vercel AI Gateway and free on OpenCode Go for two weeks.
- BlogVercelVercel: how to build end-to-end evals for agents using your API or CLIA worked method for testing software whose users are agents: a test pyramid by LLM-call count, seeded task environments, per-constraint verifiers, and choosing pass@k versus pass^k by what a failure costs.
- NewsOpenAIGPT-6 Astra helps a physicist propose a lattice definition of the Standard ModelUCSB's Nathaniel Craig says Astra helped build the construction and led the proof for a candidate nonperturbative definition, a 50-year-old open problem, now under expert review. It opens OpenAI's new science blog.
- NewsFBIFBI warns attackers are using the ARTEX agent framework to automate intrusionsThe FLASH says the open-source, LLM-driven pentest tool reached privileged access in under three hours and was used in the ConnectWise backup compromise; 1,600+ consoles are exposed. Lists IOCs and steps to lock down cloud and API access.
- NewsCursorCursor traces its breach to a leaked Fireworks key and old agent tracesAgent traces shared with a training vendor held unredacted credentials, which let the attacker reach Cursor tools and plant the fake Claude Code extension. A clear case for scrubbing secrets from agent logs before they leave your systems.
- NewsOpenAIOpenAI launches ChatGPT Wallet so agents can pay without seeing your cardBuilt with Amex, Mastercard and Visa: agents get tokenized cards and every purchase needs a security check. A concrete payments rail for agent checkout, starting with Pro users in the US.
- ReleaseDeepSeekDeepSeek API gets the app's search, files, voice, OCR, ASR and TTSBuilt-in tools switch on with a few parameters, and OCR, speech and TTS are also offered as standalone APIs, so developers get app-level features from one cheap provider.
- ReleaseHugging Facellama.cpp and ggml reach v1.0.0 with a stable API and ABIAfter 3.5 years, the core local-inference library commits to semantic versioning, so apps built on it can upgrade without surprise breakage.
- NewsGrok BotGrok Bot can now join meetingsThe bot sits in calls, answers questions, takes on assigned work and tracks team decisions, following Claude Meetings a day earlier into the same space.
- NewsGoogle ResearchGoogle opens Antigravity's /teamwork multi-agent tools after results on open problemsGoogle Research reports Gemini 4 multi-agent workflows producing results in math, formal verification, chemistry and kernel optimization, and is making /teamwork available to outside researchers.
- BlogCloudflareCloudflare sees 50+ AI pentest framework families scanning its customersForks of the withdrawn ARTEX agent now run from thousands of IPs. Cloudflare measures the traffic and explains how defenses must change for autonomous, high-volume attacks.
- BlogThinking MachinesHow Thinking Machines trained Inkling 2 Small end to end in 4-bitA detailed account of using NVFP4 in the forward pass from the first pretraining step through RL, and how they kept quantization error from hurting learning.
- BlogAnthropic ResearchAnthropic's 10 recommendations for building safety evaluationsPractical lessons on designing, running and reporting safety evals, since claims that a model is safer are only as sound as the evals behind them.
Show more
- BlogMeta EngineeringHow Meta brought Android's Cuttlefish emulator to Apple silicon MacsReality Labs ported Cuttlefish and its QEMU backend to macOS so AI-glasses developers can test against high-fidelity virtual devices instead of scarce hardware.
- ThemerisingAssistants move into the tools people already useAssistants keep absorbing the surfaces people already work in: Claude now builds whiteboards, live spreadsheets and data stories as native workspaces, and meetings became a contested surface within a day, as Claude Meetings and then Grok Bot started joining team calls to answer questions, take on work and track decisions. Muse Spaces turns a description into a persistent mini-app, and the flow also runs the other way, as OpenAI lets businesses deploy agents inside ChatGPT and Stripe exposes payments data to agents over MCP. Distribution through existing tools matters as much as raw model quality.
- SpikeNewsNVIDIAJensen Huang resigns as Nvidia CEO; CFO Colette Kress is interim CEONvidia's board says Huang has left as CEO and board member effective immediately, with Kress as interim CEO and Harvey Jones as the company's first board chair. Nvidia says its product roadmap and customer commitments continue unchanged, and Q3 results still land on Nov 17.
- ReleaseGoogle DeepMindGoogle DeepMind releases Gemini 4 Hydrogen, its fast everyday modelGoogle says Hydrogen finishes tasks 2.5x faster than Gemini 3.8 Flash and beats Argon on the Artificial Analysis Intelligence Index, with better instruction following and tool calls. Cursor reports 54.3% on CursorBench at default effort, up from 39.6% for 3.8 Flash.
- NewsAnthropicAnthropic launches Claude Meetings, Claude joins your team's video callsClaude can now sit in a team video call and build docs, whiteboards and other artifacts while people are still talking. It is rolling out in beta on Team and Enterprise plans.
- ReleaseThinking MachinesThinking Machines opens Inkling 2 Small weights under Apache 2.0The 594B-total, 17B-active MoE for agents and coding, launched last week, can now be downloaded from Hugging Face under a permissive Apache 2.0 license, so teams can self-host and fine-tune it.
- ReleaseWorld LabsWorld Labs opens its Atlas world model to everyone, with an APIAnyone can now use Atlas through World Labs' apps and API endpoints to turn an idea into a 3D world, then export keyframes, video clips or 3D Gaussian splats.
- NewsGrok BotGrok Bot on Mac can now use Messages, Mail, Notes, Calendar and ContactsThe personal agent's Mac app now works directly with Apple's built-in apps, including Messages, Mail, Notes, Reminders, Calendar, Location and Contacts, so it can act on your own data rather than only in the cloud.
- NewsAnthropicAnthropic launches its IPO roadshow with $17B quarterly revenueAnthropic opened its public offering with a prospectus showing $17.0B revenue in Q3 2026, more than 12x a year earlier, and over 11,900 customers paying $100K+ a year.
- ThemesteadyAnthropic reshapes its governance on the way to an IPOAnthropic has opened its IPO roadshow, with a prospectus showing $17.0B of revenue in Q3 2026, more than 12x a year earlier, and over 11,900 customers paying $100K+ a year. It got there after filing an S-1 to list on Nasdaq as ANT, committing 5% of its shares to a Public Benefit Fund, and handing its Agent Skills standard to the Linux Foundation's Agentic AI Foundation. A frontier lab going public puts its finances and governance in the open and sets a template for how safety-focused labs raise capital.
- NewsQwenAlibaba opens Qwen4-Max in previewThe first Qwen4 model is usable now in preview on Alibaba's Token platform, pitched at coding, front-end builds and agentic work. It is a preview, so treat it as an early look before the full Qwen4 family lands.
- NewsCursorCursor rebuilds Projects so teammates' agents work togetherProjects is now multiplayer: every teammate and their agents share one workspace and can coordinate. It moves Cursor from one developer steering agents toward whole teams of agents working side by side.
- Newsn8nn8n Labs releases Nod, an open-source personal agentNod runs on your own computer with its own browser and works across apps and sites for chores like call prep, collecting invoices or cancelling subscriptions, with the AI provider you pick. It is an open alternative to hosted personal agents.
- ReleaseGoogleGoogle launches the Gemini Decisions API with Gemini Decision 1A model that scores up to 1,000 yes/no, choice or rating questions per input in one pass, in about 200 ms at $0.10 per 1M tokens. Google now joins OpenAI in selling decision models for agent workflows.
- Newsx.comGrok Bot 1.0 brings shared bots that work with whole teamsBots can now be shared across a team, work inside Slack and Teams, and join meetings. It moves personal agents into team workflows; the launch post passed 16M views.
Show more
- ReleaseRunwayRunway releases Praxis-1, an open-weight world action model for robotsApache 2.0 weights for a model that turns video pretraining into robot control, ranked first on NVIDIA's RoboLab-120. A strong open base for teams building robot agents.
- Newsx.comDeepMind's AlphaGenome 2 predicts how DNA variants act in specific cell typesIt models how each cell type reads the same genome, so researchers can trace a genetic variant to the cells where a disease actually unfolds.
- PaperPerplexityPerplexity red-teams its agent sandbox with 12 modelsModels tried to escape the VM and bypass network policy. Findings: standard capture-the-flag setups can steer agents away from the boundary being tested, and security-tuned models did not do better.
- BlogCloudflareWhy Cloudflare chose QuePaxa, an unusual consensus algorithmCloudflare's new Meerkat control plane uses QuePaxa, a leaderless consensus algorithm, because Raft and Paxos stall when a leader slows down on the open Internet. Explains the tradeoff and how it works.
- BlogMeta EngineeringMeta's Intelligent Delivery: making every unit of ads compute countMeta's system predicts the marginal value of compute per ad request and shifts serving resources to match, adding about 10% ranking performance. Now uses agents to detect and respond to system shifts.
- NewsGoogle DeepMindDeepMind's AlphaProof Code formally verifies Signal's protocol implementationGemini-powered agents proved key security properties of Signal's compiled protocol code in four weeks, writing 6M+ lines of Lean with humans reviewing about 1%. A first for a widely deployed protocol's implementation; artifacts are open-sourced.
- ReleaseGoogle DeepMindGemini 4 Carbon opens to all paying developers at Argon's priceTwo weeks after a cyber-defenders-only preview, Carbon is in the Gemini API at $2/$10 per M tokens until Nov 30 ($4/$20 after). Google says it tops CursorBench 4.0 and Zapier AutomationBench, and Antigravity previews /swarm multi-agent runs.
- NewsClaudeClaude adds Whiteboards, Sheets and DataClaude now builds boards, live spreadsheets and data stories as native workspaces, not just chat replies. It pushes Claude further into everyday office tools; rollout started today and the post drew over 4M views.
- ReleaseGoogle DeepMindGemini Robotics 2.5 hits 96% on a Mercedes-Benz assembly line pilotDeepMind's new vision-language-action model raises success rates 30-100% across 200 tasks and ran parts kitting on Apptronik humanoids in real factories. It also open-sources ASIMOV-Jailbreak, a robot-specific jailbreak benchmark.
- NewsOpenAIOpenAI launches Businesses in ChatGPT, with Business Agents in betaBusinesses can claim their profile in ChatGPT, see how people engage with them, and build an agent that answers customers inside ChatGPT. It turns ChatGPT into a discovery channel businesses must manage, like search listings.
- NewsMuseMeta's Muse gets its own email addressPaid Muse users in the US and Canada can now have their agent send and receive email on their behalf from its own address. Another big consumer agent getting its own identity rather than borrowing the user's inbox.
- BlogEngineering at MetaHow Meta built TLX, a warp-level extension to Triton for Blackwell and HopperMeta explains why neither CUDA nor Triton fit modern GPUs and how explicit warp-group orchestration got 20-25% MFU on ads training, up to 2/3 lower RecSys inference latency, and runs MoE pretraining kernels in production.
- SpikeReleaseAnthropicAnthropic releases Claude Fable 5.2 and Claude Mythos 5.2Anthropic's top tier gets faster (up to 30%) and 20% cheaper at $8/$40 per million tokens, with a 1M context. Fable is generally available; Mythos is the same model for verified cyber and life-science orgs.
- NewsGoogleGoogle adds TPU 8tx for confidential inference and TPU 8i Max for long contextGoogle now has four eighth-gen TPU systems: 8tx keeps data and weights encrypted in use for regulated agent workloads, and 8i Max doubles 8i's compute and HBM bandwidth. It also open-sourced TPU Sync and an agent suite for Claude Code and Cursor.
- NewsDeepSeek HarnessDeepSeek Harness reports 1.54M daily users and plans a plugin marketplaceUnder three months after launch, DeepSeek's open-source agent app says it serves 1.54M daily API users and 92T tokens a day, with 35,000 community plugins. The figures are self-reported; a plugin store and author support program are next.
Startups
- StartupAll startupsLinqInfrastructure that lets AI agents text and call people over Apple Messages, RCS, SMS and voice, and use computers to finish tasks; official Apple Messages for Business provider, used by DoorDash.
- StartupAll startupsBoltzBiomolecular foundation models for small-molecule and protein design, offered through Boltz Lab to scientists and AI agents; partnered with 6 of the 20 largest pharma companies.
- StartupAll startupsPayhawkFinance operating system where AI agents handle expenses, procurement, travel, payments and reconciliation under company budgets and approvals; operates in 33 countries.