What's actually changing in AI and search, and what to do about it. Written to be read, structured so answer engines can cite it.https://nasofi.com/en-gbNabil Asofihttps://nasofi.com/blog/agent-or-workflow/https://nasofi.com/blog/agent-or-workflow/Most products sold as AI agents are fixed pipelines with a model bolted on. Run this ten-minute test before you buy or build one.Tue, 22 Sep 2026 00:00:00 GMTAn agent decides which steps to take and in what order, based on the state of the world. A workflow follows a fixed sequence with a model filling in some blanks. Ask whether removing a single step changes the path taken, not just the content produced. If the sequence never changes, you have a workflow with a language model in it — fine, but sell it and buy it as that.systemsagentic aiworkflowsautomationsystems designhttps://nasofi.com/blog/how-agent-pilots-die/https://nasofi.com/blog/how-agent-pilots-die/Most agent pilots don't fail loudly. They stall on unattributed errors, missing owners, and cost overruns nobody tracked from day one.Tue, 22 Sep 2026 00:00:00 GMTAgent pilots usually die from one of four causes: the demo-to-production gap (a clean demo hides the messy inputs production actually has), unattributed failure (nobody can say which step broke), no clear owner once the builder moves on, and cost surprise (token spend nobody modelled before launch). Each is preventable with basic logging, ownership, and a cost ceiling set before go-live, not after.systemsagentic aiproductionreliabilityai opscost managementhttps://nasofi.com/blog/what-agentic-actually-means/https://nasofi.com/blog/what-agentic-actually-means/Most products sold as agentic are deterministic pipelines with a model inside. A ten-minute test for telling the two apart, and why it changes your budget.Tue, 22 Sep 2026 00:00:00 GMTAn agentic system decides its own next step at runtime; a workflow follows steps you defined in advance. Most products sold as agentic are workflows with a language model inside, which is usually the right engineering choice. The test is whether the system can choose an action you did not enumerate — and what happens when it chooses wrong.systemsagentic aiautomationevaluationhttps://nasofi.com/blog/ai-overviews-traffic-mix/https://nasofi.com/blog/ai-overviews-traffic-mix/AI Overviews appear in roughly 45% of searches and can cut clicks to a page by more than half. Rank tracking alone cannot tell you whether that is happening to you.Wed, 16 Sep 2026 00:00:00 GMTAI Overviews appear in roughly 45% of Google searches and are associated with click reductions of up to 58% on affected queries. Impression counts fall less than clicks, so a widening gap between the two is the practical early signal. Track impressions, clicks and AI citation separately — a single ranking position no longer describes your visibility.ai-newsai overviewsmeasurementsearchhttps://nasofi.com/blog/five-workflows-i-rebuild-every-time/https://nasofi.com/blog/five-workflows-i-rebuild-every-time/Not the interesting automations — the boring ones. These five appear in nearly every engagement because they address the same handful of failure points.Tue, 08 Sep 2026 00:00:00 GMTThe five workflows that recur across most engagements are a Search Console anomaly watchdog, an inbound lead triage router, a review-mining pipeline, a content brief engine, and a feed integrity checker. They share a shape: each one watches a system that fails silently and reports the failure before a human would have noticed it.systemsautomationoperationsworkflowshttps://nasofi.com/blog/cited-not-ranked/https://nasofi.com/blog/cited-not-ranked/A page can rank fifth and get cited, or rank first and be skipped. Once you accept that, most visibility reporting stops making sense and something better replaces it.Wed, 26 Aug 2026 00:00:00 GMTRanking and citation are now separate outcomes. AI answer engines select sources for extractability, authority and consensus rather than position, so a page outside the top three can be cited while the top result is skipped. Track them as two metrics, because improving one does not reliably improve the other.ai-newsai searchcitationsstrategyhttps://nasofi.com/blog/faq-rich-results-retired/https://nasofi.com/blog/faq-rich-results-retired/FAQPage stopped producing rich results for every site on 7 May 2026. The markup is not harmful — but if it was your reason for writing FAQs, that reason is gone.Thu, 14 May 2026 00:00:00 GMTFAQPage rich results were retired for all sites on 7 May 2026, so the markup no longer produces an enhanced SERP appearance. It does not cause harm, and non-Google AI engines still parse the structure. Keep the markup if you want it, but write FAQs because your readers have the questions — not because you wanted the expanded listing.seoschemastructured datafaq