{"version":"1.0","type":"rich","provider_name":"Acast","provider_url":"https://acast.com","height":250,"width":700,"html":"<iframe src=\"https://embed.acast.com/$/659557afc7c0640016f29135/6a3bcf9ceeb75ff76e109ad7?\" frameBorder=\"0\" width=\"700\" height=\"250\"></iframe>","title":"The State of AI Engineering: What a Thousand Companies' Telemetry Reveals","thumbnail_width":200,"thumbnail_height":200,"thumbnail_url":"https://open-images.acast.com/shows/659557afc7c0640016f29135/1782304549451-7c5eb5aa-00c8-4253-a078-13d2e5edcc11.jpeg?height=200","description":"<h3>Five Moves for Leaders</h3><ol><li><strong>Adopt a model gateway</strong> — centralize routing, failover, governance</li><li><strong>Build deprecation discipline</strong> — retire models deliberately</li><li><strong>Instrument agents deeply</strong> — especially with frameworks</li><li><strong>Audit prompt caching</strong> — fix layout (stable first, dynamic later)</li><li><strong>Implement budgets &amp; backpressure</strong> — cap loops, build queues</li></ol><h3>Seven Key Takeaways</h3><ol><li>Multi-model is the norm (70%+ use 3+ models); use a gateway</li><li>LLM tech debt compounds; retire old models deliberately</li><li>Framework adoption doubled; observability burden doubled too</li><li>69% of tokens are system prompts; only 28% use caching</li><li>Context windows exploded but quality beats volume</li><li>Rate limits are the #1 failure mode</li><li>Agents are still mostly monoliths; distributed shift is coming</li></ol><h3>Key Quotes</h3><blockquote>\"The gap between a good demo and a dependable system is closed by effective evaluation and operational discipline.\" — Datadog</blockquote><blockquote>\"The next wave of agent failures won't be about what agents can't do. It'll be about what teams can't observe.\" — Guillermo Rauch, CEO, Vercel</blockquote><blockquote>\"Context quality, not volume, is the new limiting factor for LLM agents.\"</blockquote><p><br></p>","author_name":"Danar Mustafa"}