full convo is worth a watch, the hard-to-reach tactics start around 13:00:
youtube.com/watch?v=jHQDCple…
31
youtu.be/jHQDCple6DI?is=qpSf… Cameron Zoub from Whop gave some really practical (and unconventional) examples of how they go above and beyond to acquire customers in their ICP that are hard to reach like sending edible arrangements, hiring an uber driver to go knock on their house (lol), and even purchasing their product to start a convo
Congrats to the @NousResearch team on the $90M raise.
IMO it’s hands down the best agent harness for businesses and I love it.
We’ve been using Hermes agents as coworkers in Slack for months now.
Here are the 3 biggest things it’s unlocked for us:
1. Removed information bottlenecks
What did we promise this client? Why did we make that decision? Have we done something like this before?
Those questions used to suck hours of the day from multiple team members trying to triage the request.
But after we connected our Chief of Staff agent to our CRM and all our systems, now the team can ask in Slack and get the right answer in seconds.
2. We spend less time on the operational side of the work, and can just do the work....
For our content team that handles hundreds of articles a month, a single assignment can mean pulling together interview transcripts, past content, client feedback, and publication and brand requirements before even writing a single word.
Our agent can now assemble that context into briefs and drafts for the team instantly.
We still make the judgement calls. We just get to spend more time understanding the client and on the actual content. With far less mistakes along the way in terms of factual errors or missed client feedback.
3. We can upskill the agent on how our company works.
Everytime the agent makes a mistake or gets stuck, we turn the correction into a reusable skill.
How we structure a brief, what to research before recommending options in a proposal, when to ask for approvals.
As a byproduct, it's also forced us to get much clearer about our own processes.
And it means the time spent "teaching" an agent compounds. The lesson carries over to the next time the task is requested, upgrading the entire team's capabilities.
Putting our agents in Slack, where the team already works, is what unlocked agents for them.
It's no longer a siloed chat bot on your desktop or browser.
It's more like teammates that become the foundation of how the company works.
Thanks Hermes.
87
Me and my CI gates whenever I click "Merge without waiting for requirements to be met" on AI written PRs
38
Alex Karp went on CNBC yesterday and accused the frontier labs of running a lopsided trade:
you pay for tokens, they get your IP.
"The basic view among enterprises in this country is: I'm going to chillax and waste my time with tokens. I'm going to get no value, and they're going to get my IP."
Some people think he had meltdown on live tv. I think he just said what has been in the back of everyone's mind.
The labs aren't just selling picks and shovels. They're watching what everyone digs up.
Sitting at the model layer, a lab sees what no one else gets to see:
which workflows generate the most value?
which industries are adopting AI the fastest?
what do customers keep asking for and what are they willing to pay?
That strategic intelligence is worth infinitely more than the $200 subscription it comes packaged in.
It's like a heat map of every market worth entering.
I mean look at the last three months:
Claude Design launched in April. The Information reported Figma was "blindsided." Anthropic's CPO literally sat on Figma's board until three days before the launch.
Claude for Legal in May, entering the category Harvey and Legora built on top of frontier models.
Claude Science on June 30, plus Anthropic's own drug-discovery programs.
To be fair, they're starting with neglected diseases big pharma passes on. That doesn't make them a pharma company yet. But the ambition is clear: move past supplying models into making the discoveries themselves. And if you're building superintelligence, why stop at just HELPING companies?
Anthropic's operation: Watch where value is created, then move in directly.
Companies are driving spend into what they think is a model provider but is really a competitor lurking in the shadows.
And if I were them, I'd probably do the same.
It's less about villainy. More so incentive structure.
Which is why "Claude for X" isn't a moat. If your main value is using Claude better than everyone else, Claude can replicate you.
My takeaway:
The biggest risk of running your business on AI isn't that the model becomes misaligned. It's that your provider becomes your competitor.
The issue isn’t whether Anthropic or OpenAI is stealing anyone’s prompts. The issue is that they have the high ground. They see everything from above, and there's no way to verify what happens inside their walls. Ultimately, we’re relying on contracts, reputation, and incentives. That's a lot to bank your IP on.
And I say all this with my entire stack running on Claude and Codex. I mean, frick, Claude helped me with this post.
I'm going to start paying more attention to AI sovereignty.
Own your data. Own your model choices. Own where inference runs. Keep the model swappable so no lab's roadmap is existential for you.
Closed models concentrate power.
The labs aren't evil, they're rational players in a market where the dominating economic incentive is winner-takes-all.
With AI, that's potentially everything. And that's what's worrying.
1
1
188














