Steve Yegge

Former Amazon engineer Steve Yegge divided the degree to which programmers use AI into eight levels, ranging from Level 1 (not using AI at all) to Level 8 (building one's own orchestrator to direct hundreds of agents simultaneously).

In Steve Yegge's eight-level classification, Level 5 refers to completely abandoning the IDE/editor and working entirely in a terminal or conversational interface.

The 'Rule of Five' proposed by Yegge states that when using an LLM to do something, it often takes having it review its own output four to five times to reach a deliverable state.

After Yegge handed a game to AI for security hardening and the AI indicated the results were good, running Snyk again revealed 241 vulnerabilities.

Five months ago, Yegge wrote an article titled 'Software Survival 3.0,' whose core argument is that LLMs can synthesize any software but don't want to waste tokens, and therefore are happy to call tools to offload cognitive burden.

When AI increases code delivery speed by 10x, if the security defect rate remains unchanged, the attack surface also expands by 10x.

Five Eyes warned that the time for open-source models to catch up with top attack models is now measured in months rather than years, with Yegge estimating approximately six to seven months.

Yegge advises family members to agree in person on a set of code words to guard against AI impersonation scams that are highly convincing in both voice and video.

When Yegge ran Snyk on his own codebase, no private vulnerabilities beyond publicly known CVEs were found.

Steve Yegge is a former senior Google engineer and gave this talk representing Snyk at the AI Engineer conference.