Good morning, welcome to this week’s Lore Brief, your 3-minute brief of the most important moves in AI and tech.
This issue is brought to you by Factory, creators of Droid, the frontier software development agent. Available to anyone, with any model, in any IDE/terminal.
Anthropic warns recursive self-improvement is arriving faster than expected → Internal data shows Claude now accelerates AI development itself: engineers ship 8x more code per quarter, success rates on open-ended coding jumped 50 points in six months, and newer models can speed up their own training code by 52x. The company launched a dedicated institute to study the implications. Read more here →
OpenAI upgrades ChatGPT memory with “dreaming” → The new system automatically tracks what matters across conversations, understands when events are upcoming, happening, or over, and lets users review and steer what it remembers. It rolls out today to Plus and Pro users in the US with 2x more memory capacity. Read more here →
Cloudflare reports bots now exceed human internet traffic → Matthew Prince shared new Radar data showing agentic and crawler traffic has officially passed human traffic for the first time in Internet history, happening faster than anyone predicted. Read more here →
Factory Router makes AI coding more cost-effective by selecting the right model for the job:
Black Forest Labs adds Martin Scorsese as advisor:
Agentic traffic on the web has surpassed human traffic:
Matt Van Horn shares his full agentic engineering playbook:
Azeem Azhar highlights the real cost of AI agents → More than half of what companies pay when running agents is the model re-reading the same context over and over. The analysis points to memory and context-management improvements as the next big efficiency unlock. Read more here →
Peter Steinberger shares his recent “Build the thing that builds the thing” talk → The OpenClaw co-founder gave a keynote at Microsoft Build on how to design meta-tools and systems that compound productivity instead of just writing more code. Read more here →
Marc Andreessen shares why AI isn’t killing coding jobs → He highlighted a founder’s on-the-ground view that AI acts like the best junior engineer ever hired, boosting output without replacing senior talent. The piece argues the real winners will be teams that treat AI as a force multiplier. Read more here →
Google DeepMind releases Gemma 4 12B → Demis Hassabis announced the new open multimodal model that runs locally on laptops with just 16GB VRAM, already surpassing 150 million cumulative Gemma downloads. It ships under a fully open Apache 2.0 license. Read more here →
Ideogram drops 4.0, the best open image model yet → The new frontier model leads open-weight leaderboards for text rendering, layout control, and 2K resolution while giving full weights for fine-tuning and local runs. Read more here →
Microsoft AI launches seven new MAI models → Mustafa Suleyman announced MAI-Thinking-1, a 35B-active-parameter MoE reasoning model that hits 97% on AIME 2025 and 53% on SWE-Bench Pro while running on custom MAIA 200 silicon for 30% better performance per dollar than NVIDIA’s GB200. The lineup also includes strong image editing models and a 5B-parameter Code-1-Flash optimized for VS Code and Copilot. Read more here →
