@mfishbein@bhalligan > HubSpot should spend a small fortune on expert evals from Mercor. That's what takes GTM agents to the next level.
Nope, they should be buying something like SyntheticUsers or alternatives, that’s where the GTM feedback loop is
Hear me out: in 2027, OpenAI and Claude will let you build agents that run on their platforms and expose the UI/control plane embedded in their apps.
Customers will pay for your agent through the OpenAI/Claude subscription, they’ll take a commission.
It’s been a long time coming, all pieces are in place.
GLHF
PSA since it took me a second to discover it:
If you wanna get some untapped value out of your ChatGPT subscription, go into settings and turn on global dictation
It uses the same fast and accurate transcription path as the in-app dictation
the vision these people have on product management is hilarious at times. it’s all about requirements and code. Never mind the user experience, product marketing, positioning, marketing, sales enablement, support, adoption and iteration, etc. None of this will collapse into a single role, it’s just not possible. But someone has too coordinate all the planes so that something good lands in the user’s hands, in a form they can understand. 🤷♂️
@vincejaco as you well know, flying to SF isn't just pitching one fund. it's pitching tens of funds, all more or less in the same places, writing bigger checks for lower traction as long as the opportunity is real.
@khash it's quite the bother tbh, because the user-facing product and the team of bots are really good. but the underlying models are really poor, feels gpt-4o level.
never been a better time for dep supply chain attacks.
legitimate critical security updates coming out every day.
everyone's lowering their guards and disabling the cooldown period that package managers have, so they're not caught out by the legitimate updates.
@thsottiaux and Dots team, keep hitting these with my Dot. It’s becoming super frustrating because it breaks the “fire and forget” flow. I tell the Dot to do something, it says it needs my approval to do it. I have to then approve what I told it to do. Really frustrating.
It’s useless to capture everything and shove it all into an LLM. The noise is so much it negates all utility. Who cares about a micro decision made 12 months ago, even 3 months ago? There’s a reason we thankfully forget this stuff and only retain relevant memories.
i’m not convinced that “context” is the real bottleneck here.
for lots of orgs, all the meetings are captured in granola. all async comms are in slack. all docs are in notion. I’ve even seen teams check their agent logs into shared repos.
all of these have ample MCPs and CLIs that all agents can access. the context is there, everything is legible, and yet—
the problem is that it is stored, retrieved, and acted on non-deterministically. it’s unstructured and out of date and there’s way too much of it. it’s an unreliable mess. the agents (and humans) can only do so much with that, and the result isn’t at all what people imagine when they design agentic systems, because nobody can trust it.
As a product person I wouldn’t want to build stuff… I’d want to deliver stuff, give me an agentic operation, that includes product marketing, content marketing, social media management, customer success, etc. This idea that product is only for delivering features the same as a junior project manager is very myopic
@thsottiaux and Dots team, keep hitting these with my Dot. It’s becoming super frustrating because it breaks the “fire and forget” flow. I tell the Dot to do something, it says it needs my approval to do it. I have to then approve what I told it to do. Really frustrating.