9 AI Techniques You Probably Haven't Tried

Nine AI techniques are delivering strong results but remain underused. The episode draws on public sharing from LinkedIn and X rather than solely the host's direct experience, and covers tools and workflows spanning cancer research, enterprise AI adoption, voice interaction, agent management, and writing quality.

Moderna and Merck announced a successful phase three trial for a personalized mRNA cancer vaccine, the first late-stage trial to show promise. The treatment analyzes cancerous cells, identifies the DNA mutation causing the cancer, and creates a personalized mRNA vaccine to correct it, extending time in remission across more than 1100 patients with advanced melanoma. Moderna stock rose 70 percent after the announcement and reached as high as 125 percent growth within hours. A commenter cautioned that calling the treatment an AI breakthrough is misleading because it uses machine learning closer in kind to AlphaFold than to large language models, and that attributing the result solely to AI erases the human scientists involved.

OpenAI introduced private safety processing for eligible API customers with zero data retention agreements. The feature extends automated safety scanning across an entire session including multiple agentic steps, with no human ever viewing sensitive data. If an issue is flagged, an OpenAI employee receives only a summary with a category and severity rating stripped of customer data. Anthropic's response to long-horizon agent safety monitoring was instead to disable zero data retention for Claude, which has contributed to poor enterprise adoption. One observer described private safety processing as one of those small things that is actually a huge thing for enterprise AI adoption, given that enterprise customers have made commitments to their own customers requiring strong data privacy.

OpenAI also announced a partnership providing the model for Replit's newly launched FreeMode, which routes queries through GPT-4o mini Luna and allows users to create 30 times more on a normal twenty-dollar-a-month subscription without spending usage credits. The host framed this as a signal that OpenAI is competing simultaneously on the capability frontier and on the efficiency frontier against Chinese open weight models, and predicted that efficiency competition will intensify as users mature in understanding which power levels suit which tasks.

Bloomberg reported that SpaceX approached coding agent startup Cognition about a potential deal, possibly including computing capacity. Cognition most recently raised at a 2.6 billion dollar valuation in May and is reportedly seeking at least 4 billion dollars in a new round. Cognition said it is not for sale and has not been in talks with SpaceX about a deal, while Elon Musk said SpaceX only spoke with Cognition about making Grok work well for their needs. The host noted that the ambiguity of what constitutes approaching someone about a deal makes it possible that both Bloomberg's sources and Cognition's denial are accurate.

Voice mode is emerging as a genuinely different working style rather than a convenience feature. Ali K. Miller described the difference between ambient interaction and click-to-speak voice mode as sounding negligible but being a world of distance in practice. She uses Codex live voice mode on walks to fire off computer tasks, triage her day, and improve system architecture via natural language, calling it an ambient voice-controlled operating system rather than just an ambient chief of staff.

Grokbot has its own virtual computer that helps solve access issues that plagued previous agents. Use cases people are sharing include application review, sales deck updates from notes, Salesforce reports, and daily briefings. Lenny Rachitski demonstrated using an MCP containing transcripts from more than 500 of his podcast episodes to turn Grokbot into a product strategy, growth, and career advisor. Grokbot agents can already interact with one another, enabling patterns like a chief of staff Grokbot managing other Grokbots. Most agent use so far has been single-player, meaning individuals managing agents working for them rather than agents shared across teams. Claude Tag gives a Claude instance access to the entire context of a specific Slack channel and makes it a shared resource across a team, with different Claude instances in different channels able to have different permissions, tool access, and context access.

AI writing still contains recognizable tropes including self-congratulatory phrasing and the phrase that matters. Capturing things you do not want AI to do in a consolidated skill and providing it consistently is described as a better approach than repeatedly asking for rewrites. Some people are using specific style guides like the ASD-STE-100 plain writing standard and turning those into AI skills. The slash design command in Claude Code brings up an artboard workflow allowing users to leave notes, highlight areas to change, and preview different styles. Ali Miller's 18 two-word prompts treat AI interaction like addressing a team member. The prompt remember this forces AI to log a mistake or critical context error in its memory, which AI does not always do automatically, and the prompt simulate it gets AI to run scenarios, plan for edge cases, and visualize outcomes in a planning interface.