Before you let an AI agent act for you, ask: if it goes wrong, how fast can you fix it? 5 min → delegate A day → reconsider Harmful → don’t Full video is available on TwoSetAI@Youtube. @hi_angelinayang

CS freshmen asking if they just made a huge mistake. Ofer Mendelevitch: “junior engineer = new grad” isn’t a law. Graduate at mid-level instead. Full video is on TwoSetAI @ YouTube. @hi_angelinayang

Pricier model ≠ better results. Nicole Königstein found some top models burned 4x the tokens overthinking, and still lost to a small open-source model. Full interview is live on TwoSetAI@Youtube. @hi_angelinayang

Mark Hay, CTO of TextQL, on why their take-home allows AI: the skill isn’t writing code, it’s directing it and checking its work. Full video is on YouTube: TwoSetAI. @hi_angelinayang

Engineers aren’t dead — they’re moving from actors to directors. Agents write the code; you own architecture and the non-obvious calls. The full video with Ofer Mendelevitch is on TwoSetAI @ YouTube. @hi_angelinayang

CAPTCHA was built to keep bots out. Now we want agents in — and the internet still blocks them all. Full episode with Zach Meltzer of VeryAI is live on the TwoSetAI channel. @hi_angelinayang

Face ID: 1 in 1 million false acceptance rate. Two palms: 1 in 100 trillion. Zach Meltzer of VeryAI on why the palm beats the face. Full episode is live on the TwoSetAI channel. @hi_angelinayang

Gold tables for everything. A semantic layer for every column. Mark Hay, CTO of TextQL, on why overbuilding your data stack first is backwards. Now on YouTube: TwoSetAI @hi_angelinayang

Your agent knows the answer but you're not cleared to see it. Does it refuse — and leak that something's there? Or play dumb? Wiley Jones, CEO of DOSS, on the red wire problem in agent design. Full episode on the TwoSetAI channel. @hi_angelinayang

Facebook has one of the world's largest face databases — because we hand it over without thinking twice. Zach Meltzer of VeryAI on biometric privacy. Full episode is live on the TwoSetAI channel. @hi_angelinayang

AI is great at exploring, not so great on rails. Mark Hay, CTO of TextQL, on why heavy guardrails fight its strengths. Now on YouTube: TwoSetAI @hi_angelinayang

Knowledge graphs excite everyone who's never built one. If it solves 0.1% of queries, is it worth the team and upkeep? Ofer Mendelevitch on graph ROI. Full video: TwoSetAI @ YouTube. @hi_angelinayang

Trillions going into AI generation. Barely billions into AI detection. Zach Meltzer of VeryAI on why that gap is the real deepfake problem. Full conversation on the TwoSetAI channel. @hi_angelinayang

A 40-person company. No data team. Mark Hay, CTO of TextQL, on why sales, the CRO, marketing, and finance all own the data. Now on YouTube: TwoSetAI @hi_angelinayang

"Never press this button." Transcribe that and the meaning vanishes — which button? Ofer Mendelevitch on the red button problem in multimodal RAG. Full video: TwoSetAI @ YouTube. @hi_angelinayang

Today's "autonomous" agents aren't. Nicole Königstein: real autonomy means an agent can redesign itself—spot its errors, rework the architecture, cut the agent that adds nothing. Full episode on the TwoSetAI channel. @hi_angelinayang

Source says 90%. Your AI output says 85%. RAG calls it a match — because most of the words are the same. Semantic similarity isn't factual accuracy. Full video: TwoSetAI @ YouTube. @hi_angelinayang

Predicting a World Cup match is just like predicting a stock — gather player form, coaching, injuries, and reason through it. Simon Du on why LLMs act more like fundamental analysts than number-crunchers. Full video is out on TwoSetAI@Youtube. @hi_angelinayang

Vibe-coded agent passes tests, looks reasonable. A month into production, your CTO storms in over the token bill. Nicole Königstein: you can't outsource the thinking. Yet. Full video is on Youtube now. @hi_angelinayang

Trillions going into AI generation. Barely billions into AI detection. Zach Meltzer of VeryAI on why that gap is the real deepfake problem. Full video is on Youtube now. @hi_angelinayang