"Artificial intelligence is the science of making machines do things that would require intelligence if done by men."
— Marvin Minsky, Computation: Finite and Infinite Machines, Prentice-Hall, 1967
⟳ Editor's note: this edition originally published without Anthropic coverage due to a news-feed outage. The Anthropic items below were added on 2026-07-05 from the vendors' announcement archives.
The platform layer is winning: AI vendors are racing from models to durable infrastructure
Today's headlines reveal a consistent strategic pattern across OpenAI: the Ona acquisition adds persistent cloud runtimes to Codex, BBVA's 100k-seat rollout shows ChatGPT Enterprise becoming operational backbone, and Academy courses institutionalise adoption. OpenAI is not competing on model quality alone — it is building the org-level platform stack that makes switching costly. For product engineers, this means your architectural choices today (which agent runtime, which enterprise tier, which compliance tooling) are locking-in decisions with 3-5 year horizons, not 6-month experiment bets. Evaluate vendors on infrastructure depth and enterprise governance, not just benchmark scores.
| Vendor | Change | Category | Impact | Decision | Why |
|---|---|---|---|---|---|
| Anthropic | Anthropic suspends all customer access to Fable 5 and Mythos 5 following a U.S. government export-control directive. Source → | Policy / Access | Any product built on Fable 5 or Mythos 5 is immediately non-operational; builders must audit dependencies and identify fallback models now. | Watch | Government-mandated access suspension is a hard stop with no builder workaround — if you're on these models, you need a migration path today, and if you're not, this is a live reminder that export-control risk is a real architectural variable. |
| OpenAI | Acquiring Ona to give Codex secure, persistent cloud environments for long-running enterprise agents | Infrastructure / Agents | Codex moves from a code-completion tool to a durable agentic runtime — enterprise devs can now build multi-step, stateful agent pipelines on top of it | Watch | Acquisition is announced but not closed; wait for SDK/API details before committing architecture decisions to Codex agents |
| OpenAI | Codex used in production by astrophysicist Chi-kwan Chan to build black hole simulations and test GR theory | Developer Tools / Scientific Computing | Validates Codex for complex numerical / scientific code generation beyond typical web-dev use cases | Use Now | If Codex handles relativistic physics simulations, it can handle your domain-specific numerical code; experiment immediately |
| OpenAI | BBVA scaled ChatGPT Enterprise to 100,000 employees in a full banking AI transformation partnership | Enterprise Adoption | Sets a new benchmark for org-wide LLM rollout; signals OpenAI is positioning ChatGPT Enterprise as an ERP-tier platform | Watch | Study BBVA's rollout playbook for your own enterprise pilots; the 100k-seat scale reveals what change management and governance look like at the limit |
| OpenAI | OpenAI Academy launches three practical courses on AI workflows and applying agents at work | Education / Workforce | Creates a structured, free upskilling path; reduces the internal training burden for teams adopting AI tooling | Use Now | Low effort, high return — assign relevant courses to teammates before your next AI product sprint to baseline shared vocabulary |
| OpenAI | OpenAI backs the EU Code of Practice on AI content transparency and provenance standards | Compliance / Trust | EU-facing products must plan for content provenance metadata; C2PA-style watermarking will become a compliance checkbox | Watch | No enforcement yet, but architects building EU-deployed AI products should audit their content pipelines for provenance hooks now |
| Gemini Omni and Gemini 3.5 demos published, showcasing new multimodal and reasoning capabilities Source → | Model Capabilities | Gemini stack gains parity pressure on OpenAI in multimodal reasoning; dev teams using Vertex AI should re-evaluate model selection | Watch | Demos are promising but benchmark against your actual use case before switching; model capability gap between vendors is narrowing fast | |
| OpenAI / Preply | Preply ships AI-generated lesson summaries and personalised language exercises powered by OpenAI | EdTech / Product Pattern | Demonstrates a repeatable pattern: LLM layer over human expert sessions to generate personalised follow-up content at zero marginal cost | Use Now | This post-session summary + exercise generation pattern is directly portable to any product that has human-delivered sessions (coaching, consulting, tutoring) |
| Tool / Model | Category | Why It Stands Out | When to Use |
|---|---|---|---|
| OpenAI Codex (with Ona acquisition pending) | Agentic Code Execution | The Ona acquisition signals Codex is becoming a stateful cloud agent runtime, not just an autocomplete engine — the black hole simulation case confirms it handles real scientific complexity today | When you need an AI that can execute, iterate, and persist code across multi-step tasks in a secure cloud environment rather than just suggest code in an IDE |
| ChatGPT Enterprise | Enterprise AI Platform | BBVA's 100,000-seat deployment proves this is now a board-level infrastructure decision, not a productivity experiment — OpenAI is building the organisational layer, not just the model | When your org needs governed, auditable AI access across business units with SSO, usage analytics, and data privacy guarantees already baked in |
| OpenAI Academy | Developer / Team Enablement | Free, structured courses directly from the model provider that cover agent application — removes the 'where do we start' problem for teams just beginning AI adoption | Before kicking off any internal AI workflow project; use as pre-reading to get cross-functional teams speaking the same language on agents and automation |
| Experiment | Goal | Effort | Expected Outcome |
|---|---|---|---|
| Replicate the Preply post-session summary pattern in your own product | Auto-generate a personalised summary and 3 follow-up action items after any user session (support call, coaching, onboarding) using a single GPT-4o prompt with session transcript as context | Low | Demonstrable personalisation at zero marginal cost per session; creates a high-value touchpoint that increases user retention and perceived product quality |
| Run a Codex agent on your most tedious internal data pipeline script | Give Codex your ugliest ETL or data-munging script and ask it to refactor, document, and add error handling autonomously — measure how many review cycles it takes vs. a human junior dev | Medium | Baseline data for your team's Codex ROI case; reveals where agentic code execution breaks down and what guardrails you need before the Ona persistent-environment features ship |
| Type | Item | Change | Notes |
|---|---|---|---|
| Updated | OpenAI Codex | Acquiring Ona to add secure, persistent cloud environments enabling long-running agentic workflows | Acquisition announced, not yet closed; API/SDK surface changes TBD — architecture decisions should wait for GA documentation |
| Added | OpenAI Academy Courses | Three new courses on applying AI at work, building repeatable workflows, and using agents in everyday tasks | Free; suitable for non-technical stakeholders as well as engineers; useful pre-sprint team alignment resource |
| Updated | Google Gemini (Omni + 3.5) Source → | New multimodal and reasoning model variants demoed with expanded capability set | Available via Vertex AI and Google AI Studio; benchmark against your workload before migrating from existing model stack |
Get each brief in your inbox — the insight, the key updates, the verdicts. Five issues a week, every weekday.
No spam. Unsubscribe in one click. Powered by Resend.