Alan Boyd

The way people perform diligence is fundamentally shifting. Old way: Google you, click links, read, form their own opinion. New way: Ask AI. AI clicks, reads, forms an opinion, then tells them what to think. AI systems are deciding what humans believe about you. The operators who control this layer now will have an advantage that will likely be cost prohibitive to close later.

@farzyness Verified this anecdotally myself as well. Everybody thinks it's just a thing to help them write their wife's birthday card.

One of my biggest realizations from the last week or two is that 99% of the population in the US is DRAMATICALLY under-utilizing AI. Most people haven't the faintest idea that AI is no longer just a chatbot.

Had the same observation. Went through my contacts - 100s of operators - not one thought AI could do more than help them write their wife's birthday card.

@bytecrafter_1 @satyanadella Bingo. Good stuff.

The maverick slot is the part most teams skip. They build the lanes, optimize the routing, then lose the serendipity that made the multi-model approach valuable in the first place. We do something similar. One wildcard response on every critical call. The best outputs often come from the model that wasn't supposed to answer.

New in M365 Copilot: Council. You can run multiple models on the same prompt at the same time, so you can see where they align and diverge, and understand what each adds.

@bytecrafter_1 @satyanadella Yes we run this internally where each one has its own lane based on where they excel, but we also patched in a maverick response option on the back end, just like in a real war room setting where one executive can have a novel idea outside of their lane.

Every CTO I know was already doing this manually. Three tabs, three models, pick the best output. Microsoft just productized that workflow. The interesting next step is when the system learns which model wins for which task type and routes automatically. You stop comparing and start trusting the routing layer. That's where the real efficiency unlock lives.

@satyanadella We built the same thing internally and call it our AI Charter

@clear20ai @lydiahallie It was a fresh chat thread, simple question like whats on the agenda today?

@alanjboyd @lydiahallie What was this “simple question” and how did it use 10k+ tokens in 30 seconds? Likely you’re suffering from one or more of these: - peak hour throttling - massive context req. to start - long chat history compounds context window size - cache window (only 5 min for pro)

We're aware people are hitting usage limits in Claude Code way faster than expected. Actively investigating, will share more when we have an update!