LLM answers can't be traced, so they can't enter audited enterprise systems
DavidLinthicum · x · 2026-09-26
Accuracy gets the attention, but traceability decides whether an LLM ever reaches production.
- David Linthicum says every time he tries to trace how an LLM got to a bad answer, he can't — and untraceable output can't go into a business system that gets audited
- The take came out of a discussion with Eric Kavanagh and David Sweenor on Inside Analysis (Sept 18), an episode framed as "the AI slowdown already happened inside your company"
- Sweenor's write-up argues enterprise AI adoption has stalled largely because human-in-the-loop review keeps models stuck: outputs that can't be explained or audited can't be shipped
- The trio continues the conversation in person at the Agentic Road Show in Pittsburgh on Oct 22
The practical takeaway for buyers: compliance and auditability, not leaderboard scores, often decide which models make it into production.
More from Companies & People
- GrowthXAI founder's birthday essay: from immigrant cafeteria to building belonging — santilli · 2026-09-26
- Meta may expand Muse to 1B users, each agent sandbox costing ~$0.33/hour at market rates — sergeykarayev · 2026-09-26
- X rolls out first payouts under its new Original Content Rewards program — deobfuscations · 2026-09-26
- Chinese firm AI adoption is low but accelerated after DeepSeek, NBER study finds — avicgoldfarb · 2026-09-26
- Microsoft drops Copilot+ PC branding — new Surface PCs qualify but won't carry the name — Ars Technica AI · 2026-09-26
- Another Google DeepMind researcher quits, calls building superintelligence 'inherently irresponsible' — The Decoder · 2026-09-26