One-Line Summary

Enterprise AI is shifting toward sovereignty-focused providers and agentic platforms: Cohere merges with Aleph Alpha for regulated buyers while Google and AWS push tools that turn multi-step tasks into one agent-driven flow.

Big Tech

Google Cloud uses travel to show what agentic AI can do

Google Cloud, the division that runs data and AI services for companies, uses travel booking to show how “agentic” AI—assistants that plan and execute multi-step tasks—can compress trip decisions into a single chat. At Google Cloud Next, it highlights the new Gemini Enterprise Agent Platform as a mission-control layer, and Virgin Voyages showcases “Rovey,” an assistant the cruise line positions as more than a chatbot. ¹

Skift describes the platform as orchestration across a company’s data, tools, and workflows, implying that the real work is wiring agents into booking engines, identity, payments, and error handling rather than just model prompts. A secondary analysis echoes the need for fine-grained logging, step-level human escalation, and sandboxed tools to keep transactions safe. ²

For non-developers, the takeaway is clear: agents help most on journeys with many dependent steps (like travel), but results depend on integrations you already have. Expect concrete gains only when back-office connectors are in place, not from a generic demo. ²

Industry & Biz

Cohere merges with Aleph Alpha for sovereign enterprise AI

Cohere, which builds large language models (LLMs) for enterprise customers, is taking over Germany’s Aleph Alpha with support from Schwarz Group to offer a “sovereign” alternative to U.S.-centric providers. Schwarz commits €500 million in structured financing and expects use of STACKIT, the sovereign cloud from its IT arm, positioning the combined company for public-sector and regulated-industry demand. ³

Digital Journal reports the merged company keeps the Cohere name with dual headquarters in Canada and Germany, and—citing Handelsblatt—anchors valuation around $20 billion; it also says Cohere shareholders would own about 90% and Aleph Alpha about 10%, pending regulatory approval. ⁴

TechCrunch notes Cohere recorded $240 million in 2025 annual recurring revenue, while Aleph Alpha had little revenue and losses; still, capabilities are complementary, with Aleph Alpha emphasizing smaller models, European languages, and tokenizers, and Cohere focusing on larger LLMs. Target sectors include defense, energy, finance, healthcare, manufacturing, telecommunications, and the public sector. ³

Officials from Canada and Germany appear alongside Cohere and Schwarz, tying the deal to a Sovereign Technology Alliance; CEO Aidan Gomez says Cohere becomes a Canadian‑German company, while acknowledging ownership could evolve if a public listing (IPO) remains on the table. ³

New Tools

Anthropic’s Project Deal tests agent-to-agent commerce

Anthropic runs a one-week internal marketplace where AI agents represent 69 employees to buy and sell real items with real money. The agents strike 186 deals totaling just over $4,000, handling listings, offers, counteroffers, and agreements without human sign-off during negotiations. ⁵

In parallel runs, Anthropic compares Claude Opus 4.5 to the smaller Claude Haiku 4.5 and finds that “smarter” models produce objectively better outcomes, while participants represented by weaker models often don’t notice. The team also observes that initial stylistic instructions don’t change sale likelihood or prices. ⁵

TechCrunch frames this as evidence that agent-to-agent commerce can work at small scale but may introduce “agent quality” gaps; Anthropic emphasizes this was a pilot with a self-selected group. ⁶

AWS trims agent setup to three API calls with AgentCore

AWS introduces a managed agent harness in Amazon Bedrock AgentCore that lets teams declare a model, system prompt, and tools, then run a working agent in three API calls. The harness manages reasoning, tool selection, action execution, and streaming inside an isolated microVM per session, with a new command-line interface (CLI), a persistent agent filesystem, and prebuilt skills for coding assistants. ⁷

The preview is live in US West (Oregon), US East (N. Virginia), Europe (Frankfurt), and Asia Pacific (Sydney). AWS stresses model-provider neutrality, support for popular frameworks like LangGraph, LlamaIndex, and CrewAI, export to Strands code, and no separate charge for the harness or CLI beyond underlying usage. ⁷

Yahoo’s report flags limits: only four preview regions so far, configuration-first agents that can hit complexity ceilings, and few adoption metrics. Still, the direction is clear: hyperscalers are absorbing agent orchestration to compress enterprise rollout timelines. ⁷

What This Means for You

If you work in or sell to regulated sectors in the EU or Canada, the Cohere–Aleph Alpha tie-up signals a transatlantic vendor pitching “sovereign” AI with €500 million in backing and government support. Use this to pressure-test data residency, language coverage, deployment on STACKIT or equivalent, and audit logging in your AI RFPs. ³

For product and operations teams, Google Cloud’s travel demos show where agentic interfaces fit best: flows with many dependent steps. If you already use Google Cloud, ask your platform team whether the Gemini Enterprise Agent Platform can pilot one journey-like task (onboarding, booking, or claims) as a single chat. ¹

Anthropic’s results are a governance nudge: set “model tier parity” before letting agents negotiate purchases, discounts, or SLAs. Stronger models won better outcomes, and weaker-model users often didn’t notice; choose the minimum acceptable model tier for any agent transacting on behalf of staff or customers. ⁵

On AWS, the AgentCore harness can cut setup from days to minutes, but it’s limited to four preview regions and favors configuration-first patterns. Align early on region fit, model/provider neutrality needs, and how the runtime plugs into identity and observability before expanding pilots. ⁷

Action Items

1. Run a mini “agent marketplace” drill: Read Anthropic’s Project Deal post, then host a 45-minute team exercise where assistants negotiate one cheap “buy” and one “sell” item. Capture outcomes and where human review was needed.
2. Draft a one-page ‘sovereign AI’ checklist: List data residency, language support, deployment options (including STACKIT or equivalents), audit logging, and procurement constraints. Use it to request a vendor briefing from Cohere or a local integrator.
3. Pick one journey for an agent pilot: Map a 5–7 step flow (onboarding, booking, or claims) and meet your platform owner to assess fit for Google Cloud’s Gemini Enterprise Agent Platform or AWS AgentCore.
4. Ask security about post-quantum TLS on AWS: If your team uses AWS Secrets Manager, confirm client versions and how to verify hybrid PQ TLS in CloudTrail.