@okikusan-public
A personal tech blog.
Notes from the intersection of AI × work × product.
Where I write up what I've tried in the field, what I've thought through, what I've learned. AI product adoption, internal rollout, the Second-Brain pipeline — themes adjacent to my day job (AI solutions consulting / SE).
▍ HIRO KIKUCHI Latest post
[ latest / 7 ]From editor to agent management — Google Antigravity 2.0 marks the arrival of the Agent OS
Antigravity 2.0 is not an editor — it is an Agent OS. Not the same layer as Claude Code / Codex / Grok Build, but the layer binding Desktop / CLI / SDK / API. The comparison axis moves from "which model is smarter" to harness design / permission boundaries / context / scheduling / review.
Don't build an AI that replays yesterday's spec — the gap between spec and source of truth is the real context
What separates AI agents in 2026 is not the model or the prompt — it is whether you accumulate the gap between spec and Source of Truth, and the reasons behind the drift. The blind spot of Spec-Driven Development and the real core of Harness Engineering, laid out.
AI agents enter the territory code can't write — long-tail × tacit knowledge × tacit thoughts
Automation has always stopped at explicit knowledge. AI agents are the first technology to descend into tacit knowledge and 'tacit thoughts' — the situation-aware territory. Connecting Nikkei xTech's 'long-tail × AI,' Polanyi, Nonaka's SECI, and Karpathy's LLM Wiki into one layered model.
Hermes Agent — The execution engine for your Second Brain
Nous Research's OSS self-improving agent. A closed loop of task → skill → memory delivers "teach once, never forgets." It turns Obsidian's static Second Brain into a dynamic execution agent.
Don't ask which job AI replaces — ask which tasks inside the job
Microsoft Suleyman's 12-18 month prediction. What gets replaced isn't job titles — it's information-processing tasks done at a computer. What stays: judgment, accountability, connection. Why Applied Engineer / FDE gains value.
Beyond "Deployed but Unused" — Designing Internal AI Adoption
Copilot rolled out company-wide but only 10% use it. ChatGPT Enterprise contracts that don't open once a month. Design the natural adoption loop with AI Champions × Monthly LT × Hackathons × KPI alignment — from both the buyer and provider side.
Obsidian → LLM Wiki → HTML → AI Deploy
Just keep dropping rough notes — the AI grows a wiki behind the scenes, writes HTML in JA / EN, and ships it to a static host for you. A note on speeding up second-brain publishing with AI.
→ See all articles