MondAI Roundup - June 2026

MondAI Roundup has been presented by Dominik Lukeš since late 2024. It gives an overview of AI news from the previous month and links them to key themes of interest. It runs every second Monday of every month at 12:30–13:30. A regular summary of key lessons and a more detailed write-up live on AI News Roundup.

June did not have one announcement that made everything feel different overnight. It was more like the changes we have been watching since the start of the year kept accumulating. Microsoft Build put Microsoft IQ, Work IQ, Fabric IQ, Scout and governed agents into one enterprise stack. OpenAI's updated ChatGPT memory did the same context work from the consumer side, with a background process that synthesises useful context from previous conversations.

The centre of the practical tool section was agent workflows. Anthropic explained Claude Code dynamic workflows as task-specific harnesses that can coordinate subagents. OpenAI kept moving Codex toward general knowledge work, with mobile control, plugins, Sites, app context and workflow features. Google Antigravity 2.0 now looks much closer to the same Codex-like shape, and Google AI Studio is getting mobile apps and a handoff into Antigravity.

The model section was more mixed than a simple frontier-benchmark list. Microsoft launched seven in-house MAI models. MiniMax M3 pushed the open-weight frontier. Gemma 4 12B mattered because it can run locally on powerful consumer hardware and handle multimodal input. Google Eloquent then showed the practical product version of that direction: local dictation and voice editing on Mac and iOS.

The business section kept returning to compute. GitHub Copilot's move toward consumption pricing, Anthropic's Claude limits and Agent SDK accounting, Ramp's enterprise AI spending data, and NVIDIA/Microsoft local AI hardware all point to the same constraint: better agents run longer, use more tools, and cost more to operate. The later sections covered security, science and public institutions: Cloudflare and Mythos, Project Glasswing, OpenAI Rosalind, coding agents in social-science research, Google's AI Co-Scientist, maths results from OpenAI and DeepMind, Pope Leo XIV's AI encyclical, Olga Tokarczuk, and Pangram's AI-detection debate.

For more details, read the detailed notes or browse the presentation slides.

Key Stories

• Microsoft Build and workplace context — Microsoft IQ, Work IQ, Fabric IQ and Scout put workplace context, governed agents and Copilot into one enterprise frame.
• ChatGPT memory and dreaming — OpenAI rebuilt memory around background context synthesis and reviewable user context.
• Claude Code dynamic workflows — Anthropic described agent workflows that can create task-specific harnesses, coordinate subagents and run longer work.
• Codex for knowledge work and Codex workflow features — OpenAI is positioning Codex as a work environment for reports, spreadsheets, presentations, contracts, research and automation.
• Google Antigravity 2.0 and Google AI Studio mobile — Google's agent tools are converging on mobile control and agent-first development workflows.
• Microsoft MAI models, MiniMax M3 and Gemma 4 12B — the model section mixed in-house product models, open weights and local multimodal deployment.
• Google Eloquent, Microsoft MAI Image 2.5, Ideogram 4 and Reve 2 — voice and image work is moving toward editing, layout and product integration.
• GitHub Copilot consumption pricing, Claude Agent SDK accounting and Microsoft's RTX Spark Dev Box — pricing and hardware now shape what agents can do in practice.
• Cloudflare/Mythos, Project Glasswing, OpenAI Rosalind and Google AI Co-Scientist — security and research workflows are becoming serious AI deployment sites.
• Pangram and AI detection, Olga Tokarczuk and AI and Magnifica Humanitas — public institutions are trying to separate sensible AI use from suspicion, enforcement and status anxiety.

Resources

• Detailed notes — expanded chapters based on the session transcript
• Presentation slides
• Quick reference index — searchable list of people, organisations and links

Key Links

• Microsoft Build 2026
• ChatGPT memory and dreaming
• Claude Code dynamic workflows
• Codex for knowledge work
• Google Antigravity 2.0
• Gemma 4 12B
• Google Eloquent
• GitHub Copilot consumption pricing
• Cloudflare on frontier cyber models
• Pangram policing is the new grammar

Next Session

Monday 13 July 2026, 12:30 BST — register here. Same time, second Monday of the month, same Teams.