<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Miguel Oliveira</title><description>Writing on machine learning, AI engineering, and adjacent things.</description><link>https://miguelcabraloliveira.com/</link><item><title>Notes on the future: AI, SaaS, and the markets that will bleed</title><link>https://miguelcabraloliveira.com/blog/notes-on-the-future/</link><guid isPermaLink="true">https://miguelcabraloliveira.com/blog/notes-on-the-future/</guid><description>AI won&apos;t kill SaaS. It will compress per-seat pricing, bleed single-dev DevTools, and bleed early-stage technical VC. The rest of the market just changes shape.</description><pubDate>Sat, 30 May 2026 00:00:00 GMT</pubDate></item><item><title>Agent harnesses don&apos;t transfer</title><link>https://miguelcabraloliveira.com/blog/agent-harnesses-dont-transfer/</link><guid isPermaLink="true">https://miguelcabraloliveira.com/blog/agent-harnesses-dont-transfer/</guid><description>Everyone now agrees that the harness contributes more than the model. What people are not saying out loud is that the harness pattern we converged on was shaped for coding, and copying it into other domains is making most non-coding agent products worse than the chatbots they were going to replace.</description><pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Claude Council: when one agent isn&apos;t enough</title><link>https://miguelcabraloliveira.com/blog/claude-council/</link><guid isPermaLink="true">https://miguelcabraloliveira.com/blog/claude-council/</guid><description>A Saturday hack that turns Claude Code skills into a multi-agent debate. The chat interface is the wrong shape for hard decisions.</description><pubDate>Sun, 15 Feb 2026 00:00:00 GMT</pubDate></item><item><title>Engagement surveys are theatre</title><link>https://miguelcabraloliveira.com/blog/engagement-surveys-are-theatre/</link><guid isPermaLink="true">https://miguelcabraloliveira.com/blog/engagement-surveys-are-theatre/</guid><description>We used to do qualitative research. Then scale broke it and we built surveys instead. LLMs let us go back, if we get the pipeline right.</description><pubDate>Wed, 08 Oct 2025 00:00:00 GMT</pubDate></item></channel></rss>