@muratcan I think most people still have the opposite experience with models.
How do you think about something like the Remote Labor Index? Humans ranked models as matching human freelancer work quality 16% of the time.
@KSimback Spot on. This insight is so undervalued. Agents shouldn't just be productivity body doubles for employees. There's way more to get out of them if they have the right context.
@conconconling Agree. Not convinced FDEs is the answer, but people who don't actually talk to mid-market and enterprise leaders aren't grasping the size of the opportunity.
@RamaswmySridhar Agree not all companies need a full knowledge factory, but most need more than just shared definitions.
Shared terminology on top of obsolete info, process intel with major gaps or contradictions, etc. doesn't get agents very far.
Tons of companies are building “context layer” tools to give agents information that will supposedly help them perform better.
We decided to test what type of context makes a difference and found that more =/= better.
I wrote about the growing context layer market, what info agents actually need (and don't), and what our internal benchmarking showed.
@eglyman Super interesting. Consistent with what I hear talking to companies to. Small cohorts of power users are finding new ways to use AI, but almost everyone else has settled for low-hanging use cases and aren't exploring more or pushing the limits as models improve.
@sattyyouneed Owning the data models can't train on. General knowledge is a commodity. Companies produce tons of proprietary data that the labs would kill for but that most companies do nothing with. Capture that.
Great essay.
"Ultimately, what we’re doing is mistaking adoption for transformation. And the biggest gains will not be from making existing organizations more productive, but from designing new kinds of organizations that couldn't have existed before."
Absolutely fantastic piece by my partner @edavidpeterson. I'm biased, but I think it's a must read.
"The search for the AI-native firm"
https://t.co/CYE4pEC0JU
Hosted a @Scribe dinner last week w/finance, HR, marketing, HR, ad operations leaders to discuss closing the AI value gap. I asked: “what do you wish AI could do that it can’t today?”
Here's what I heard 👇
5. People want honesty about how little most companies have changed. X and LinkedIn can make it look like everyone else's AI work is lightyears ahead of yours. We talked about the elephant in the room... that for the majority companies, AI hasn't meaningfully changed how they work at all.
@scribe 4. People want AI to teach you how to use it better. An L&D leader pointed out that if it takes 5-10 prompts to get you to a good answer, AI should tell you how to get the same result in fewer prompts next time.
@scribe 3. People want data-backed AI roadmaps. Automating the wrong thing is expensive, and people want clearer ways to make big bets without it feeling like guesswork.
@scribe 2. Demos are way ahead of delivery. Multiple people mentioned being wowed in demos and then never seeing ROI in practice and wanting to close that gap.
@scribe 1. Nobody wished for smarter models. Leaders recognize the tech is already there and the issue for their agents is the context the models don’t have.
@dennisihong Solid advice. I'm most bullish on proprietary data and systems of record/embedded workflows. For software businesses, those are the biggest slice of what's left.
I noticed that folks with kids are better at using AI agents. Probably because they have lots of experience in instructing someone intelligent who looks for loopholes in order to claim that a task was accomplished.
@SherryYanJiang Yes, in theory, but "slow down" doesn't create understanding.
People and companies don't understand their work bc they don't have the visibility to and work changes constantly. Slow down to do what?