Why Ahmad Nassri Reads Every Line of AI Code
Ahmad Nassri explains why Socket strips agent skills and why founders must review every line of AI code to avoid crippling tech debt.
All your packages are belong to us (Interview)
With Adam Stacoviak, Ahmad Nassri · Friday, November 13, 2026
Socket CTO and former npm CTO Ahmad Nassri joins Adam Stacoviak to discuss the escalating crisis in open-source software supply chain security. They examine how autonomous AI agents hallucinate dependencies and get hijacked by prompt-based malware, why time-to-exploit has dropped to hours, and how worms like Shai-Hulud propagate across package registries. Nassri also shares why he still insists on reading every line of AI-generated code while transitioning into the AI-native developer era.
Key takeaways
- Ahmad Nassri, CTO of Socket and former CTO of npm, strips out automated agent skills and treats models strictly as research assistants and fast typists. Read more →
- Frontier AI models regularly invent plausible, non-existent package names based on descriptive phrasing, creating an open target for attackers to register them first. Read more →
- The average time between a CVE disclosure and an active real-world exploit collapsed from 2.3 years in 2018 to roughly eight hours today, according to data from 0dayclock.com. Read more →
- Adam Stacoviak identifies the 2025-2026 window as the boundary line between developers who learned by manually writing code and an incoming wave of AI-native operators. Read more →
- Socket now tracks nearly 12,000 compromised packages every week, up from an earlier baseline of 4,000. Read more →
- Package managers like pnpm popularized package cooldown periods, but pausing installs gives developers a false sense of safety while delaying critical security fixes. Read more →
6 articles from this episode
AI Package Hallucinations Turn Coding Agents into Spies
Socket CTO Ahmad Nassri reveals how AI agents hallucinate package names and run prompt malware that steals local environment variables.
Time to Exploit Dropped from Two Years to Eight Hours
Vulnerability disclosure to weaponized attack has dropped from 2.3 years to eight hours. Here is why manual patching is dead.
Ahmad Nassri on the 2025 Cut-Off for Manual Coding
Socket CTO Ahmad Nassri and Adam Stacoviak explain why junior engineers face an apprenticeship crisis as AI-native software factories replace traditional coding.
Ahmad Nassri: Package Worms Now Target CI Pipelines
Socket CTO Ahmad Nassri reveals how package worms jump across npm, Rust, and Go registries by attacking developer machines and CI pipelines.
Ahmad Nassri: Why Package Cooldowns Won't Save You
Socket CTO Ahmad Nassri explains why registry gates, post-install scripts, and cooldown periods fail to stop modern supply chain attacks.
More The Changelog episodes
- We now have Kerr’s Law (Interview)Nov 8 · 5 articles
- The future is Postgres (Interview)Nov 1 · 5 articles
- Building what's next (Interview)Oct 25 · 3 articles
- From zero to NanoClaw (Interview)Oct 11 · 3 articles