Eric Faust

@useapolloio @tempo Yep! Great for basics, but you can’t tune skills and chain steps quite as well as in CLI. Harness can also do much more once your list is built on the GTM side 👀 Plus, can build monster lists outside constraints of subscriptions and tokens with MPP.

@ehfaust @tempo you taken our claude MCP for a spin yet? 👀

We turned one market thesis into 905 enriched contacts across 256 accounts. Not by buying another GTM subscription. Not by spending a day clicking filters in Apollo or Clay. Claude Code did the GTM engineering. @useapolloio supplied the data. @tempo / MPP gave us usage-based access. This is where GTM is going.

GTM is becoming programmable. Filters are the old interface. Agent harnesses are the new one. DM for early access to our harness.

This is why we are releasing a local GTM agent harness that turns a natural-language market attack into: - target companies - enriched contacts - ranked accounts - rejection logic - attack cohorts - messaging angles - campaign-ready exports No bloated stack required.

Open source won developer infrastructure. Agent harnesses may be different. Generalist open source harnesses are great for experiments. But commercial agents need to be accessible for the average person. ... and commercial agent harnesses are starting to win big time.

The next SaaS pricing model is becoming obvious: seats + agent consumption. Seats still matter for identity, access, and permissions. But agents turn one user into hundreds of actions. The companies that make their APIs headless enough to meter that work are going to print.

Imagine if Google did not rank results. It just returned a massive A-Z "marketplace" of websites. That is where agent commerce is right now. Agents do not need more API directories. They need one ranked answer: which service should I call for this task? That is the missing layer. That is Nitrograph.

@abdulalali Exactly. The hard part isn’t just finding services, it’s getting from intent to a call that works. The problem to solve is a probabilistic agent vs a deterministic endpoint. The interesting moat might be the memory of what actually worked. Best example of problem:

discoverability for agentic-services is increasingly, table stakes. core advancement is in the lens of bridging the gap from intent to execution: (1) this involves providing discoverability based on context, user/agent intent, and memory of preferences (2) involves surfacing the best "tools," for discoverability in the. form of services - evolving to the potential for chaining tools to drive towards an output (3) services involved in #2 (agents or tools) will eventually have elements of "defensibility," due to a combination of moats surrounding usage (i.e data and graph moats, discoverability/preference ratio from agents, and ability to be ranked higher due to reputation)

@MagaShawn Hey @MagaShawn definitely think you're on the right track here. When you say AgentPay settles it are you referring to Mastercards AP spec? The repo you shared is private so can't take a peak.

@ehfaust This is exactly the gap we've been building toward. We just shipped the x402 open spec — EIP-712 signed grants, a 5-step payment flow, conformance test vectors, and a reputation layer — precisely so agents can call, pay, and report outcomes from the same workflow you're describing. The "Gotchas" layer you're naming maps directly to what we formalize in specs/grants.md — pre-flight metadata an agent needs before money moves. Nitrograph finds the service. x402 grants authorize the call. AgentPay settles it. That's the full stack. Repo: github.com/shawnhvac/x402 — would love to be the reference implementation Nitrograph points to for payment execution.

@brendan_j_ryan Would love to chat with the Tempo team about enabling Gotcha Cards for MPP. The pain point is that on first call, agents often have to infer how to use an endpoint from ambiguous metadata. They end up learning the call pattern probabilistically. Example here:

@brendan_j_ryan Yep, for MPP we’re indexing from the official discovery directory/spec endpoint, currently mpp.dev/api/services. We refresh it nightly, store the full service JSON including endpoints/payment metadata, embed it for semantic search, and expose it through Nitrograph.

I live in my terminal all day running agents. So I don’t want to browse API directories. I want to tell my agent what I need and have it find the right service. Here I asked it: “Use Nitrograph to find the best lead gen services.” Nitrograph returned high-confidence x402/MPP services directly in the CLI.