Need to know
- Anthropic ships Claude Opus 4.8 with Dynamic Workflow, orchestrating hundreds of parallel agents for large-scale coding tasks.
- OpenAI launches a biodefense program, opening model access to organizations working on biological risk mitigation.
- KPMG embeds Claude across all 276,000 employees in 138 countries, starting with Tax and Legal.
- Okta shares surge 18% after Q1 beat, with analysts raising price targets to $120 on AI identity security demand.
- SentinelOne cuts 10% of its workforce despite crossing $1B ARR, signaling a profitability squeeze in crowded EDR market.
New Releases
Anthropic released Claude Opus 4.8 today, with the headline feature being Dynamic Workflow: Claude Code can now spawn and orchestrate hundreds of parallel subagents to execute complex engineering tasks simultaneously.
- Demonstrated capability: porting a 750,000-line Bun codebase from Zig to Rust in 11 days with a 99.8% test pass rate, using parallelized subagents.
- Available now via the Anthropic API, Claude Code, and GitHub Copilot for eligible users; pricing holds at the same rate as Opus 4.7, with a 15x premium request multiplier in Copilot before usage-based billing.
Google DeepMind released Gemini Embedding 2 today, a native multimodal embedding model available through the Gemini API and Google Cloud Vertex AI.
- Supports retrieval across text, image, video, audio, documents, and code in a single model, removing the need to stitch together modality-specific embedding pipelines.
- Enterprise relevance: available on Vertex AI immediately, making it a direct option for teams building RAG pipelines on Google Cloud infrastructure.
Palo Alto Networks unveiled Idira today, a next-generation identity security platform targeting human, machine, and AI-agent identities, arriving shortly after the company completed its acquisition of CyberArk.
- Timed to the CyberArk integration: Palo Alto closed the roughly $25B CyberArk acquisition earlier in 2026, and Idira appears to be the first combined identity product surfacing from that deal.
- AI-agent identity is the explicit pitch: the product is framed around securing non-human identities as AI adoption accelerates, putting it in direct competition with CyberArk's legacy PAM installed base now under PANW ownership.
OpenAI announced a biodefense program today, granting expanded model access to organizations working on detecting and countering biological risks.
- Selective access model: the program provides elevated API capabilities to vetted biodefense researchers and organizations, carving out a policy exception to OpenAI's standard usage restrictions around biological content.
- Strategic positioning: the launch comes as Anthropic has been loudly associated with Mythos-class cybersecurity capabilities, giving OpenAI a visible biosecurity narrative to counter.
Funding
Cohere acquired German sovereign AI company Aleph Alpha to create a combined entity valued at roughly $20B, positioning the pair as the primary non-US alternative to OpenAI and Anthropic for European enterprises with data residency requirements.
Asana paid $75M for no-code AI agent platform StackAI, a direct bet that the future of project management is orchestrating agents across enterprise systems rather than tracking human tasks.
OpenAI launched a $4B consulting subsidiary backed by TPG, Goldman Sachs, McKinsey, and Capgemini, acquiring Tomoro for 150 day-one engineers, signaling OpenAI's move into professional services to capture enterprise deployment margin it has been ceding to SIs.
Case Studies
KPMG has embedded Claude into its Digital Gateway platform for all 276,000 employees across 138 countries, with Tax and Legal going live first and full rollout targeted for September 2026.
- Scale and sequencing: Tax and Legal are first movers because they generate the highest per-hour billable value and carry the most structured, document-heavy workflows suited to LLM augmentation.
- Named deployment, hard timeline: the September 2026 full-rollout date and the specific employee and country counts make this one of the largest confirmed enterprise Claude deployments on record.
Okta reported Q1 revenue of $765M, up 11.2% year over year with EPS of $0.91 versus a $0.72 consensus, driving an 18% single-day stock rally as analysts raised price targets to $120.
- AI identity is the growth narrative: management cited securing AI agents as the primary demand driver, with multiple analysts noting it as the reason for the raised targets from both Needham and Truist Financial.
- Stock reaction reflects forward confidence: the 18% move on an 11% revenue growth beat suggests the market is pricing in accelerating growth from non-human identity workloads, not just current execution.
Trending on X
- Anthropic nearing $1T valuation Multiple accounts are circulating data showing Anthropic surpassing OpenAI in US business adoption per Ramp spending data (34.4% vs 32.3%), with valuation discussion pushing toward $1T and sparking debate about whether the enterprise market has effectively bifurcated between the two labs.
- Agentic coding token cost anxiety Ethan Mollick and others are noting that despite CFO anxiety over token spend, no organization that has adopted AI coding is retreating, and the real challenge is figuring out which token expenditures represent building versus which are retry tax from poor context management.
- Microsoft new coding model incoming Street Insider and multiple accounts are circulating a report from The Information that Microsoft will release a new coding-focused AI model next week, generating speculation about whether it challenges Cursor and GitHub Copilot's current positioning.
- GPT-5.5 Instant quietly ships OpenAI's Greg Brockman confirmed GPT-5.5 Instant in ChatGPT today with no formal announcement, prompting discussion about whether OpenAI is shifting to a continuous release cadence that bypasses the traditional launch cycle.
- LLM vs agent framing debate A thread distinguishing LLMs as the brain from AI agents as brain-plus-harness is gaining traction, with practitioners arguing that most enterprise teams are still deploying LLMs and calling them agents, creating accountability gaps when tasks fail.