Personal archive / ongoing
A quiet calendar for the releases, ideas, decisions, and small turns that may matter later.
Archive calendar
| MON | TUE | WED | THU | FRI | SAT | SUN |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | |||
| 5 | 6 | 7 | 8 | 9 | 10 | 11 |
| 12 | 13 | 14 | 15 | 16 | 17 | 18 |
| 19 | 20 | 21 | 22 | 23 | 24 | 25 |
| 26 | 27 | 28 | 29 | 30 | 31 | |
The list
Each entry stays intentionally light: a date, a title, and a note on why it deserves to be remembered.
Personal agents like OpenAI's Dots are the right long-term goal, but now is the wrong time. This essay argues that we should build Agentic Software first, and shows how Sign in with ChatGPT and Claude Mods help.
Intelligence is getting cheaper faster than electricity or compute ever did. Using Xiaomi's open-source MiMo post-training code as an example, this essay shows why the agent era needs a new system for knowledge collaboration.
Xiaomi released and open-sourced the MiMo-V2.6 Pro and Flash multimodal models, alongside their technical report and resources for reinforcement-learning research.
OpenAI released GPT-6 Sol for complex coding and agentic workflows, with text and image input through the Responses and Chat Completions APIs.
Anthropic released Claude Opus 5.5 for agentic coding and knowledge work, with stronger performance and lower typical running costs than Opus 5.
DeepSeek released DeepSeek-V4.1-Flash, a native multimodal Mixture-of-Experts model that pairs a causal encoder-decoder architecture with a one-million-token context window and a substantially smaller KV cache.
How Agentic Software carries intelligence into real-world workflows—and how verified experience returns as knowledge, better Agents, and a path toward RSI.
OpenAI released GPT-6 Astra, its most capable model for end-to-end work across computer use, coding, research, and professional workflows, beginning with a limited trusted-access rollout.
Anthropic released Claude Fable 5.1 for demanding reasoning and long-horizon agentic work, strengthening coding, multistep research, and document-heavy workflows.
Z.ai released GLM-5.3, a post-training-driven update focused on stronger coding, longer agent tasks, and newly observed cybersecurity capabilities.
As agents enter every software environment, the missing layer is not another chat interface. It is a governed way for people and agents to collect, review, transfer, and reuse knowledge through real work.
A new milestone in models. The question to revisit later: did it transform long-horizon agent work, or only improve benchmark scores?
Kimi released its 2.8-trillion-parameter K3 model with native vision, a one-million-token context window, and a focus on long-horizon coding and knowledge work.
Z.ai released GLM-5.2 for long-horizon tasks, pairing a one-million-token context window with stronger project-scale coding and sustained agent execution.
Agentic AI matters not simply because it makes software more intelligent, but because it turns the ability to operate on knowledge—once supplied only by people—into a productive input that can be called, organized, verified, and continuously improved.