A new podcast about AI agents from @nevermined_io and @NapthaAI. Hosted by @dongossen, @richardblythman, and @HoodAndLedger
Spotify →
Joined February 2025
- Tweets102
- Following5
- Followers109
- Likes45
ONESHOTTED
@dongossen @HoodAndLedger @richardblythman
@NapthaAI @Nevermined_ai
08/22/2025
We delve (ahem) into if we have, indeed, been oneshotted, by @_opencv_'s definition.
But hey, who hasn't?
Major CEOs, politicians, the White House, even Randy from South Park got got by LLMs.
Maybe the conversation needs to be about being more sophisticated about how you use them (we explained how we're using them) ?
Maybe... we're doomed.
The enterprise market put on their nicest pair of khakis and hit the Las Vegas Strip for AI4. So did we.
In this episode:
– AI4 field notes: 10+ tracks, big attendance, lots of intro-level “agents for enterprise ROI.” More vendor lock-in than open networks, more consulting than product.
– The monopoly vs bespoke question: When codegen + cheap tooling make custom builds trivial, does Salesforce-style dominance still hold?
– Consulting in disguise: “Agent startups” that are really frameworks + billable hours. Why observability claims mostly mean dashboards, not real runtime truth.
– Launch corner: Stack Bench: a pragmatic way to test how coding agents actually use your library (semantic vs agentic doc search, deprecated calls, evals that matter).
– Docs as product: “AI-readable” docs, and CI-style updates so your agent interface doesn’t rot.
– Tool selection is dumb (for now): ChatGPT’s agent UX is molasses; true “pick the right tool from infinity” is the ASI tell.
– Model attachment is real: version deprecations create GPT psychosis.
– Where interfaces go next
00:00 Fun and Easy Banter about Vegas
01:49 AI4 Conference Experience
03:27 Enterprise AI Implementation Challenges
07:30 IT Consulting and Agent Development
14:02 Stack Bench Launch
21:01 GPT Interface Issues
23:26 GPT-4 Access and AI Relationships
24:43 Future of Human Communication
🎙️@HoodAndLedger, @richardblythman, @dongossen
GPT-5’s biggest change isn’t in a benchmark.
It is routing. You no longer pick a model. The platform routes your request.
That one decision moves the center of gravity from “which LLM” to “which workflow.”
In this episode we unpack what mixture-of-models means for cost, latency, product design, and why agents become the app layer as models converge.
We also get practical: where GPT-5 actually helped, how Gmail/Docs integrations might collide with platform politics, and gave some founder mode updates.
🎙️@HoodAndLedger @dongossen @richardblythman
We're back.
After a few weeks hacking and yapping everyday with startups/major AI Labs in both San Francisco and China, @dongossen, @richardblythman, and @HoodAndLedger feel pretty well informed on the current state of AI agents.
On the pod this week we discussed the aversion to anything crypto in AI, MCP vs. Browser-based agents, and speculated on the upcoming GPT-5 release (is it AGI?)
Check it out ↓
Last week, @dongossen @richardblythman and @HoodAndLedger discussed the current positioning moves we are seeing from all the model providers, why the ultra premium tier is so popular, and why the enterprise is lagging in AI transformation.
Plus... could an AI dot com bubble be in play?
Find out ↓
Welcome back to Trillion Agents.
This week: a dose of reality for the crypto AI crowd.
Everyone’s building infra.
No one’s shipping products.
And the excuse is always the same: “we just need better marketing.”
But maybe it’s not a marketing problem. Maybe it’s a market problem.
@HoodAndLedger is fresh off ETHCC and calls out the elephant in the room: empty panels, big promises, and zero applications in sight.
@dongossen calls BS on the go-to-market excuses.
@richardblythman breaks down why multi-agent systems are being built in San Francisco, not onchain.
If Web3 wants to play in the trillion agent future, it needs more than pure belief.
Let’s get into it.
The great @richardblythman has returned!
He joined @Nevermined_io cofounder @dongossen to discuss the current trajectory of the AI agent space, they touched on @karpathy's latest context engineering vs. prompt engineering banger, did some rapid fire AI agent ideas, and more.
Enjoy!
We're back! At least 2/3's of us..
While @richardblythman remains in an undisclosed (hopefully tropical) location, @dongossen and @HoodAndLedger hopped on the pod to discuss one of the AI agent space's most controversial topics:
Outcome-based pricing
So far, companies have priced agents in a variety of ways. From baking them into their existing enterprise SaaS model to charging as a function of usage.
As of late, outcome-based pricing has become a popular option especially with the consumers of agentic services.
This naturally makes a lot of sense.
"You only pay when the system is successful" is an easy pitch.
Not only does outcome-based pricing satisfy consumer psychology, it also protects against the simple fact that AI agent failure rate is still quite high.
So this sort of pricing model may make a lot of sense..right now.
...When competition is low and failure rate is high.
On an office hours style podcast of sorts we discussed the rise and (in our opinion) impending fall of outcome-based pricing as the preferred model for AI agents.
If you're in the weeds, building these systems this should hopefully be a valuable perspective.
Enjoy
Discover your ai agent's true cost
xitter.cf/Nevermined_io/status/1…
Still guessing what to charge for your AI agent?
We built a solution: The Nevermined AI Pricing Calculator
Here’s how it works:
1️⃣Pick a use case – Chat, media generation, research…
2️⃣Customize components – models, tools, etc.
3️⃣Set your target margin
✅In seconds you’ll get:
– Cost per interaction
– Recommended Price
– Revenue and profit projections
Try it for free!
🔗: pages.nevermined.io/pricing-…
