<?xml version='1.0' encoding='UTF-8'?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
<channel>
<title>Toltec Systems engineering notes</title>
<link>https://www.toltec.io/engineering-notes.html</link>
<description>Engineering notes from Toltec Systems: lessons from three decades of infrastructure, applied to AI, security and software.</description>
<language>en-gb</language>
<atom:link href="https://www.toltec.io/feed.xml" rel="self" type="application/rss+xml"/>
<lastBuildDate>Wed, 28 Oct 2026 16:13:43 +0000</lastBuildDate>
<item><title>Reusable agent skills. Controlled execution.</title><link>https://www.toltec.io/note-ai-agent-skills.html</link><guid isPermaLink="true">https://www.toltec.io/note-ai-agent-skills.html</guid><pubDate>Wed, 28 Oct 2026 09:00:00 +0000</pubDate><category>AI engineering</category><description>What belongs in an agent skill, what the tools must enforce and how to review, test and operate reusable AI workflows.</description></item><item><title>Change the model. Recheck the work.</title><link>https://www.toltec.io/note-ai-model-changes.html</link><guid isPermaLink="true">https://www.toltec.io/note-ai-model-changes.html</guid><pubDate>Wed, 28 Oct 2026 09:00:00 +0000</pubDate><category>AI engineering</category><description>A practical release method for AI model changes: compare real tasks, check outcomes and permissions, inspect failures and keep a route back.</description></item><item><title>When the sources disagree, show the disagreement.</title><link>https://www.toltec.io/note-ai-conflicting-sources.html</link><guid isPermaLink="true">https://www.toltec.io/note-ai-conflicting-sources.html</guid><pubDate>Wed, 28 Oct 2026 09:00:00 +0000</pubDate><category>AI engineering</category><description>How to handle conflicting, outdated or inaccessible evidence in AI retrieval: source authority, effective dates, permission checks and a clear unresolved state.</description></item><item><title>The agent stopped. Did the action happen?</title><link>https://www.toltec.io/note-ai-agent-recovery.html</link><guid isPermaLink="true">https://www.toltec.io/note-ai-agent-recovery.html</guid><pubDate>Wed, 28 Oct 2026 09:00:00 +0000</pubDate><category>AI engineering</category><description>Recovering interrupted AI workflows without blind retries: durable action identity, confirmed state, bounded retries, renewed approval and compensating actions.</description></item><item><title>Prove parity before you rewrite.</title><link>https://www.toltec.io/note-prove-parity.html</link><guid isPermaLink="true">https://www.toltec.io/note-prove-parity.html</guid><pubDate>Sun, 04 Oct 2026 09:00:00 +0000</pubDate><category>Engineering</category><description>Before rebuilding a legacy system, capture what it actually does as tests, then replace it in slices behind a facade. Where AI helps, and where it does not.</description></item><item><title>RAID to retrieval. Failure modes that never changed.</title><link>https://www.toltec.io/note-raid-to-retrieval.html</link><guid isPermaLink="true">https://www.toltec.io/note-raid-to-retrieval.html</guid><pubDate>Fri, 25 Sep 2026 09:00:00 +0000</pubDate><category>Engineering</category><description>Storage and AI retrieval need different integrity checks. Lessons from RAID about silent failures, controlled rebuilds and checking evidence end to end.</description></item><item><title>What NetWare taught us about AI agent permissions.</title><link>https://www.toltec.io/note-netware-agent-permissions.html</link><guid isPermaLink="true">https://www.toltec.io/note-netware-agent-permissions.html</guid><pubDate>Fri, 25 Sep 2026 09:00:00 +0000</pubDate><category>Engineering</category><description>Lessons from NetWare effective rights for Microsoft 365 agents: trace each identity, permission and resource boundary to establish what the agent can reach.</description></item><item><title>Capacity planning, from Hyper-V clusters to model routing.</title><link>https://www.toltec.io/note-capacity-planning.html</link><guid isPermaLink="true">https://www.toltec.io/note-capacity-planning.html</guid><pubDate>Fri, 25 Sep 2026 09:00:00 +0000</pubDate><category>Engineering</category><description>N+1 thinking, headroom and graceful degradation kept virtual clusters running. The same discipline keeps AI systems fast and affordable under load.</description></item><item><title>What a DR test should actually prove.</title><link>https://www.toltec.io/note-dr-testing.html</link><guid isPermaLink="true">https://www.toltec.io/note-dr-testing.html</guid><pubDate>Wed, 23 Sep 2026 09:00:00 +0000</pubDate><category>Engineering</category><description>A disaster recovery test should prove five things with a stopwatch running: the data, the order, the access, the time and the people. How to test each one.</description></item><item><title>Least privilege for AI agents in Microsoft 365.</title><link>https://www.toltec.io/note-ai-agent-access.html</link><guid isPermaLink="true">https://www.toltec.io/note-ai-agent-access.html</guid><pubDate>Wed, 23 Sep 2026 09:00:00 +0000</pubDate><category>AI engineering</category><description>Current identity and access choices for Microsoft 365 AI agents: accountable identities, scoped permissions, human approval, prompt-injection controls and verified shutdown.</description></item></channel>
</rss>
