Together in agentic dreams
Montfort’s AI 3–60 Newsletter / Edition: #59 15 May 2026
Introduction
A rapprochement between xAI and Anthropic wasn’t quite on our 2026 bingo card, but here we are, with the latter becoming the former’s compute customer. Meanwhile, OpenAI has released its answer to Mythos and Claude gains additional agentic prowess in the Managed Agents platform, leveraging ‘dreaming’ among other exciting features. By way of lessons, we have a news tool from Semafor and a new App Store from Hugging Face for the robot hobbyist.
Claude dreams of outcomes
OpenAI unveils Daybreak
Anthropic partners with…xAI?
3 minute lesson: Robot Apps for All
60 minute lesson: Semafor Intelligence and the AI Summit
🔎 The Week’s Big AI News at a Glance
🔗 Claude? You’re dreaming – At the Claude for Developers conference, Anthropic unveiled some new capabilities for its Managed Agents platform. Chief among the capabilities is ‘dreaming’ where sessions and memories are reviewed and curated with the intention of learning across tasks and agents, improving performance. Outcomes, meanwhile, allows a rubric to be defined for successful task completion, with the agent defining the means and a grader evaluating success. Finally, multi agent orchestration permits a lead agent to spawn subagents with their own model, prompt and tools to reduce and conquer tasks. We think:🧠 Managed Agents abstracts away some of the complexities of running agents, including infrastructure. Add to that the ability to take Claude plugins (bundles of agent skills, connectors and files) and give them to agents and there’s a clear path from desktop chat interface to more agentic behaviour.
🔗 OpenAI’s answer to Mythos – OpenAI has unveiled Daybreak, a cybersecurity solution which teams GPT-5.5 models with the Codex Security app for finding vulnerabilities and recommending fixes. We think:🧠 Model access ladders from default GPT-5.5 models to models which have additional safeguards for working on higher risk workflows, which tend to require application and validation by OpenAI. The feature parity wars continue.
🔗 xAI’s new compute customer is Anthropic – xAI’s new customer is Anthropic, which has bought all the compute that the former’s Colossus 1 data centre can offer. We think:🧠 Techcrunch offers here a sceptic’s take on the deal, suggesting that xAI may be eyeing becoming a neocloud (leasing out, rather than using their own compute capacity to train Grok). What’s interesting is that we are seeing here two companies (with xAI looking likely to be rolled up in SpaceX) preparing for significant IPOs. For Anthropic, compute capacity was looking to be an issue imposing headroom on usage. For xAI, the emphasis is on ensuring viable use for the large-scale infrastructure the company has built out. What’s more, OpenAI’s advantage over Anthropic was in compute; no longer.
⏲️ Spend 3 Minutes With… (Quick Wins)
🔗 App Store but make it robots – when some of us were growing up, messing around with Arduino and Raspberry Pi and automation was a fast track to seeing your backpack magically teleport on a bus stop roof. Not any more. ML platform Hugging Face’s hobbyist robot Reachy Mini now has an App Store and an agentic model that can help you develop hobbyist apps.
🕙 Spend an Hour With (Deeper Dives)
🔗 Semafor Intelligence – Semafor used AI to analyse the agreements and disagreements at a high level summit on AI. The content itself is interesting, but the approach whereby AI was used to map the views of 300 speakers without seeking to offload human insight is a novel example of augmentation.
🚀 Final Thought
Our Research, Data, and Analytics team helps companies across sectors understand how to protect and enhance their reputations using advanced analytics and machine learning. We offer workshops for communications teams looking to get ahead, as well as tools and techniques to navigate AI’s impact on reputation. If you’re interested, get in touch at digital@montfort.london.