So a few weeks back I started subscribing my @hellodeckai assistant to a ton more stuff than I knew was possible. And I've been honestly shocked at how on top of it I am.
For instance, we use @HubSpot for our CRM. It keeps track of how many calls, meetings booked, customer status. Prior to Deck it took hourrrrs to chunk through it for my Monday meetings.
I created a scheduled task to benchmark our performance and deliver it in an Excel, and it's incredible.
Models are trained on the average of everything, so they gently pull all of us toward the middle of it. When, just naturally, the minority that are truly 1/100 specialists are phenomenal at something (e.g., sales, product decisions) we are optimizing for the other 99. Being the outlier can be harder than it seems.
the fact this was vibe coded in ~30 min with codex is insane. i love jazz and often ask for deep cuts from chatgpt. now i just have an incredible DJ playing tracks across whatever genre. in classic winamp skin, of course. the times we live in.
there's a weird thing happening: the work AI does best is exactly the work juniors used to learn on. so the bottom rung of a lot of careers is vanishing right as people are trying to climb it, and nobody really knows how you get good at the job without the years of grunt work that used to come first.
the tension that is happening in real-time: AI drops the cost of producing work close to zero and does nothing to the cost of deciding whether the work is good. so the load shifts onto your most senior people, who end up reviewing output all day instead of making it (which was the fun part to begin with)
Tried to use @claudeai to scrape my data from a product I’m using and paying for. Was in a back and forth debate about whether I have permission to do this even though I’m clearly the admin on the account. Switched models immediately.
we're at the point where AI's effect on firm/company output outside of the engineering dept is limited far more by people not knowing how to use a model to run ambiguous, multi-step jobs reliably than by how smart or powerful the models are
An initial startup idea can't usually be both grand and precise. In practice they're usually either grand and vague or precise and small. Precise and small is better. You know who your initial users are, and you expand outward. With grand and vague you can't even get started.
for knowledge work, email is where the assistant should live. the threads, the context, the people you're working with, it's all already there. no new app to check, no window to babysit. that's the bet behind assistants on @hellodeckai .
If AI shopping works, a lot of ecommerce brands will lose the chance to charm the customer on-site. The product data has to do more of the selling before the shopper ever sees the brand.
at what point does a better model stop mattering for most work? it feels like each new frontier release helps a shrinking slice of tasks that actually needed the extra headroom. will more powerful models == more enterprise adoption?
Zero-click commerce sounds abstract until you imagine the customer never landing on your site. They ask an assistant, it compares options, and the purchase starts somewhere else. The storefront becomes a data source as much as a destination.