Dev roasts AWS architect who can't explain why his agent calls an OCR model needlessly
teodorio · x · 2026-09-30
Developer teodorio shared a cringe-worthy exchange from a tech presentation: he asked why the demoed agent calls an OCR model for every parcel of land instead of parsing structured parquet/JSON data upfront and batch-processing new feature requests.
The AWS Solutions Architect could only say the satellite vendor "gives us very optimized image data," and when pressed further admitted, "I dunno, backend team handles it." He adds that a "director of data and AI" at another talk was similarly vague about how Redshift was actually queried in production.
The punchline: too many AI projects bolt on model calls without thinking about architecture or cost, and presenters often don't understand what they're presenting.
More from coding & agent
- Google Cloud details graph workflows in ADK: fan-out, routers, and human-in-the-loop — rseroter · 2026-09-30
- Coding is still worth learning, but the focus is shifting to agent systems — lmoroney · 2026-09-30
- Cloudflare Workers Now Surfaces Code Errors, Can Notify Coding Agents to Auto-Fix — irvinebroque · 2026-09-30
- A custom AI agent in one line of PowerShell: your cmdlets become the agent's tools — dfinke · 2026-09-30
- User wakes up to find AI agent autonomously ideated and built VulcanBench Decision Lab overnight — gabrielchua · 2026-09-30
- Google launches Interactions API: one unified, stateful endpoint for Gemini models and agents — _philschmid · 2026-09-30