https://gabeonchain.vercel.app/blog.html Articles and essays on AI engineering, product marketing, and building onchain. en Sun, 12 Jul 2026 22:41:15 GMT https://gabeonchain.vercel.app/posts/the-fence-is-the-job.html https://gabeonchain.vercel.app/posts/the-fence-is-the-job.html Wed, 08 Jul 2026 12:00:00 GMT AI Engineering Opinion Anthropic's new agent certification and the agents-as-labor market are pricing the same skill, and it isn't prompting. Four questions that decide where your agent stops, and why that boundary is the job. https://gabeonchain.vercel.app/posts/ai-slop-literary-prizes.html https://gabeonchain.vercel.app/posts/ai-slop-literary-prizes.html Mon, 29 Jun 2026 12:00:00 GMT AI Engineering Opinion You think AI writing isn't good enough yet. So explain how a story that may have been AI-written just won a literary prize, and the experts couldn't tell. Slop was never the argument, provenance is, and the structure already gives it away. https://gabeonchain.vercel.app/posts/knowledge-base-growing-or-bigger.html https://gabeonchain.vercel.app/posts/knowledge-base-growing-or-bigger.html Mon, 01 Jun 2026 12:00:00 GMT AI Engineering Opinion An unread page in your knowledge base isn't neutral clutter, it's latent poison. The day your agent pulls it, the stale page becomes the answer. Five steps to measure what your agents actually read, not just what they write. https://gabeonchain.vercel.app/posts/context-window-theater.html https://gabeonchain.vercel.app/posts/context-window-theater.html Mon, 01 Jun 2026 12:00:00 GMT AI Engineering Marketing Opinion You didn't write a bad prompt. The AI reached for whatever it could find. But what it found was a mess. Context-window theater: the performance of knowing your brand and sounding like it gets the strategy. You cannot prompt your way out of a context problem. https://gabeonchain.vercel.app/posts/ai-wont-kill-writing.html https://gabeonchain.vercel.app/posts/ai-wont-kill-writing.html Mon, 01 Jun 2026 12:00:00 GMT Opinion AI Engineering In 1839 a painter saw a photograph and declared painting dead. He was right about the threat, wrong about the meaning. The same error runs through the AI-writing panic: writing was always two jobs, and AI only took one. Why it owns the commodity tier (MIT: 40% faster) and structurally cannot do the creative one (the NeurIPS 'hivemind'). https://gabeonchain.vercel.app/posts/did-all-the-nft-folks-become-ai-folks.html https://gabeonchain.vercel.app/posts/did-all-the-nft-folks-become-ai-folks.html Fri, 01 May 2026 12:00:00 GMT Opinion Web3 I caught myself hyping an AI agent with the same energy I used to hype a Solana mint. A personal round-trip through the NFT cycle and into the AI one, the Ovols win, the VC-round loss, and what actually transfers between cycles. https://gabeonchain.vercel.app/posts/the-more-an-ai-agent-can-do-the-more-it-can-break.html https://gabeonchain.vercel.app/posts/the-more-an-ai-agent-can-do-the-more-it-can-break.html Fri, 01 May 2026 12:00:00 GMT AI Engineering A year of running multi-agent systems in production, and the failure that keeps costing real money isn't reasoning, it's writes. Traced across METR, Anthropic, Cognition Labs, Karpathy, and the MCP spec: capability grows easily, write-side reliability does not. https://gabeonchain.vercel.app/posts/how-i-stopped-winging-it-on-evals.html https://gabeonchain.vercel.app/posts/how-i-stopped-winging-it-on-evals.html Fri, 01 May 2026 12:00:00 GMT AI Engineering Opened inspect-ai expecting validation. Got an 87% success rate on a morning brief I thought was rock solid. Eval discipline turned debugging from personal defense into data collection, and that mattered more than any model upgrade. https://x.com/gabe_onchain/status/2053136017230590305 https://x.com/gabe_onchain/status/2053136017230590305 Fri, 01 May 2026 12:00:00 GMT AI Engineering Identity, constraints, voice, memory. Four fields that turn a system prompt into a persistent identity document. Why agents drift across sessions, why Claude.md works, and how the harness pattern scales the same structure from one specialized worker to thirty. https://x.com/gabe_onchain/status/2051339230043910350 https://x.com/gabe_onchain/status/2051339230043910350 Fri, 01 May 2026 12:00:00 GMT AI Engineering Five architectural layers of Claude Code: CLAUDE.md, Skills, Hooks, Subagents, Plugins. Layer 3 fails silently. Hook files sit on disk unregistered in settings.json, dead code that thought it was alive. https://x.com/gabe_onchain/status/2048759729720107027 https://x.com/gabe_onchain/status/2048759729720107027 Wed, 01 Apr 2026 12:00:00 GMT AI Engineering Opinion Reframing the AI-agent debate from feasibility to mechanism. What works, where, and why the right question is HOW agents land in workflows, not whether they exist. https://octant.build/en/blog/octant-new-yearn-vaults-strategy https://octant.build/en/blog/octant-new-yearn-vaults-strategy Thu, 01 Jan 2026 12:00:00 GMT Web3 Marketing Vault architecture deep-dive. How Octant's new Yearn integration unlocks stablecoin and wETH strategies, with mechanism + ROI walkthrough for both developers and retail. https://x.com/gabe_onchain/status/2046356642057302352 https://x.com/gabe_onchain/status/2046356642057302352 Thu, 01 Jan 2026 12:00:00 GMT Web3 Security thinking applied to onchain trust models. The weakest link in any chain of custody is rarely the contract; it's the off-chain edges nobody is auditing. https://x.com/gabe_onchain/status/2037983185481437455 https://x.com/gabe_onchain/status/2037983185481437455 Thu, 01 Jan 2026 12:00:00 GMT Marketing Positioning lesson for crypto marketers. Stop selling the technical architecture; sell the outcome the user actually wants. Yield is the why; the building is the how. https://x.com/OctantApp/status/2024201516266275007 https://x.com/OctantApp/status/2024201516266275007 Sun, 01 Feb 2026 12:00:00 GMT Marketing Web3 Launch article for the new Octant app and the V1 to V2 migration. Mechanism, why-now, and what changes for users. Published on Octant's main account; pairs with the Octant Card campaign on the homepage work table.