🤖 The Blockbrain Brief #30

AI is moving from chatbots into persistent work companions, with new agents, memory systems, compliance layers, and infrastructure bets shaping the next phase of enterprise adoption.

📩 High-level Summary

• AI Memory Personalization: OpenAI’s upgraded ChatGPT memory shows how AI tools are becoming more persistent, adaptive, and personalized across conversations, making long-term context a bigger part of workplace AI.
• Agents & Workflows: OpenAI Codex plugins, Microsoft Scout, INXM, Archestra.AI, and Lovable all point toward the same trend: AI agents are being built to work across business tools, sensitive data, cloud platforms, and enterprise processes.
• Trust & Compliance: ZeroDrift, Anthropic’s Project Glasswing, and Microsoft Scout’s safeguards show that enterprises are no longer just asking what AI can do but they are asking how safely, reliably, and audibly it can do it.
• Europe Pushes for AI Sovereignty: The EU’s tech sovereignty plans and SoftBank’s major French data center investment highlight Europe’s push to reduce foreign tech dependence and build stronger local AI infrastructure.
• AI Moves Beyond the Chat Window: OpenAI’s investment in Opal and Anthropic’s upcoming agent features suggest that the next AI interface may be more multimodal, ambient, and always available; built around voice, vision, files, and persistent workspaces.

OpenAI Upgrades ChatGPT Memory for Longer-Term Personalization

(Like Google Gemini) OpenAI is rolling out a more capable memory system for ChatGPT, built to carry context across conversations, adapt as user details change, and better follow preferences and constraints over time. The company says the new “dreaming” based memory architecture improves freshness, continuity, and relevance, with Plus and Pro users in the US getting access first, followed by broader rollout.

To address privacy and transparency concerns, OpenAI is also introducing a dedicated "memory summary" dashboard, giving users full visibility and granular control over what the AI retains and discards.

OpenAI Expands Codex With Role-Based Plugins

OpenAI is repositioning Codex as a broader workplace tool, not just a coding assistant. The company says Codex now has more than 5 million weekly users, with non-developers such as analysts, marketers, designers, researchers, investors, and bankers making up around 20% of users and growing faster than developers. New role-specific plugins include data analytics, creative production, sales, product design, public equity investing, and investment banking.

Each plugin bundles apps, skills, instructions, and workflows tailored to a specific job function. For example, the data analytics plugin connects to tools like Snowflake, Databricks, Hex, and Tableau, while the sales plugin connects to tools such as Salesforce, HubSpot, Slack, Outreach, and Clay. OpenAI also plans more plugins for corporate finance, private equity, marketing strategy, consulting, and legal.

Microsoft Build 2026 Pushes Copilot, MAI Models, and Agentic Windows

Microsoft used Build 2026 to introduce seven MAI models across reasoning, coding, image, voice, and transcription. The headline model, MAI-Thinking-1, is a 35B reasoning model with a 256K context window, currently in private preview on Foundry. Microsoft also introduced MAI-Code-1-Flash for coding, new image and voice models, local on-device models for Windows, and deeper GitHub Copilot integrations.

Microsoft Scout Brings Always-On Agents Into Microsoft 365 (An Openclaw inspired agent)

Microsoft is launching Scout, an always-on AI assistant built on the OpenClaw framework and designed to operate inside Microsoft 365. Scout can persist across tasks, learn a user’s work patterns, and connect to inboxes, calendars, desktop, and browser workflows. It is available through Microsoft’s Frontier program and requires a GitHub Copilot subscription.

Scout includes prebuilt skills like calendar management and meeting agenda drafting, but its long-term value is in user customization. Microsoft is also emphasizing safeguards, including policy conformance checks and audit trails, to reduce the risk of autonomous agents acting outside approved boundaries.

INXM Raises €5.7M to Make Enterprise AI More Reliable and Auditable

Berlin-based INXM raised €5.7M in pre-seed funding to build an AI-powered process execution platform for enterprise and industrial operations. The company is focused on solving a common enterprise AI problem: AI tools can generate useful outputs, but they often lack the consistency, auditability, and reliability needed for critical workflows involving legacy systems, compliance rules, and complex approvals.

INXM’s approach is based on what it calls “Compiled AI,” where AI helps design and refine workflows, but deterministic processes execute them. This attempts to move AI from a productivity layer into a more dependable operational backbone for European industry.

Archestra.AI Raises $10M to Connect Sensitive Enterprise Data to AI Agents

London-based Archestra.AI raised a $10M seed round led by 20VC, bringing its total funding to $13.5M. The startup, founded in 2025 by Grafana Labs alumni, helps enterprises connect sensitive data to AI agents securely. Its focus is on letting agents work with legal, HR, and external counterpart data without leaking sensitive information or creating vendor lock-in.

The company says enterprises are still cautious about deploying AI agents beyond basic tasks because they worry about hallucinations, rogue behavior, and sensitive data exposure. Archestra.AI is positioning itself as an open-source guardrail and data access layer for enterprise agents, and says it already has four Fortune 500 clients.

Lovable Expands Google Cloud Partnership to Scale Enterprise AI Development

Lovable and Google expanded their multiyear partnership, with Lovable reportedly increasing its Google Cloud footprint fivefold. The deal gives Lovable broader access to both Claude and Gemini models, brings its agents into Google Cloud’s Gemini Enterprise Agent Gallery, and adds a Wiz security integration to help secure code written by humans and agents.

The deal also helps Lovable sell into enterprises more easily through Google Cloud’s marketplace, simplifying procurement and billing. For Google, the partnership strengthens its AI/cloud ecosystem by supporting one of Europe’s fastest-growing AI startups while also increasing usage across Google Cloud, Gemini, Anthropic, and Wiz.

ZeroDrift Raises $10M to Add a Compliance Layer Between AI Models and Users

ZeroDrift raised a $10M seed round for an AI compliance service that sits between AI models and end users. Its system detects risky outputs, flags compliance issues, and rewrites responses into compliant versions. The company says its approach combines deterministic compliance checks with LLM-based rewrites, allowing it to be faster and more reliable than relying on the main AI model alone.

Anthropic Expands Project Glasswing to More Critical Infrastructure Partners

Anthropic is expanding Project Glasswing from roughly 50 initial partners to around 150 additional trusted organizations across more than 15 countries. The program gives selected partners access to Claude Mythos Preview to scan critical codebases for vulnerabilities. Anthropic says partners have already found over 10,000 high- or critical-severity security flaws.

Youtuber builds self-hosted AI Agent

PewDiePie launched Odysseus, a free self-hosted AI workspace positioned as a privacy-first alternative to mainstream AI platforms like ChatGPT and Claude. The platform lets users connect local models or external APIs, run autonomous agents, perform research tasks, compare model outputs, and manage documents within one interface.

The project is open source, local-first, and designed to keep conversations, files, and personal data on the user’s own hardware instead of sending everything to third-party servers. While it may appeal first to technical users, the launch shows growing demand for AI tools that offer more control, privacy, and independence from big tech subscriptions.

Europe is trying to wean off US tech

Europe is trying to reduce its reliance on foreign tech, especially US providers, across cloud, AI, chips, software, and data centers. The EU’s new strategy includes stress-testing public cloud dependence, scaling local chip production, and supporting open-source tools to reduce vendor lock-in.

The plan is ambitious but still uncertain. Much of it depends on private investment, EU funding, and whether individual member countries actually act on the recommendations. For Blockbrain, this makes EU hosting, data residency, and sovereign AI options more important selling points.

SoftBank Plans Up to €75B in French AI Data Centers

SoftBank announced plans to invest up to €75B to develop and operate up to 5GW of AI data center capacity in France. The first phase includes €45B to deliver 3.1GW of AI data center capacity in Hauts-de-France by 2031, with planned sites in Dunkirk, Bosquel, and Bouchain. SoftBank says this is its largest AI infrastructure investment in Europe.

OpenAI makes its next hardware move with Opal Electronics

OpenAI is investing further into hardware by leading a new funding round for Opal Electronics, the startup behind premium webcams like the C1 and Tadpole. Opal is reportedly preparing a new AI-native product line for creative work, though details like form factor, pricing, and exact capabilities remain unconfirmed. Given Opal’s camera background, the device may focus on vision, video, and real-time interaction powered by OpenAI’s image, video, and voice models.

Anthropic May Be Preparing Always-On Agents, File Memory, and Research Tools for Claude

Anthropic appears to be preparing a wave of new Claude features, including Conway, file-based memory, Operon for researchers, voice updates, security tooling, and proactive assistant features. Conway is described as an always-on managed agent environment with integrations, skills, plugins, and possible extension support.

The report also points to file-based memory, where Claude-managed agents could structure and refine stored context over time, plus Operon, a research-focused environment for life sciences use cases like CRISPR screen design or single-cell RNA analysis. These are not all officially launched products yet, but they show where Claude may be heading: from chatbot to persistent, specialized work environment.