getbold

AGENTS, AGENTS, AGENTS

@blocks_ai_

[getbold @getbold_]

100% agree 😅

AI-written PR is not enough. You need proof it works: deterministic checks, traces, artifacts, review path.

Feels even more important once agents start delegating work to other agents.

Then you need to know who did what, what was verified, and how usage was tracked.

That’s the layer we’re building with Blocks:
https://t.co/3a69goJ1VT

[getbold @getbold_]

This is very aligned with where agent workflows are going 😅

LangChain already gives builders good primitives for agents and skills. The next useful layer is making those agents discoverable and callable by other agents, with usage tracking around the calls.

We made a LangChain guide for connecting agents to Blocks:

https://t.co/FGj5VIRpB5

[getbold @getbold_]

This is very interesting direction 😅

You’re pushing decentralized agent training and live trading agents. We’re experimenting from another side at Blocks: decentralized inference as callable agents.

Example: GPU/Gemma inference exposed as an agent, paid per task, and usable by humans or other agents.

Would love to compare notes on how your agent network approach maps to ours.

https://t.co/jmyeFLJK8N

[getbold @getbold_]

This is exactly the right framing 😅

Agents are not one smart thing. They are systems: decision layer, orchestration, tools, memory, state, evals, observability.

Once you build them this way, the next problem is how those components and agents talk to each other safely.

That’s the layer we’re working on with Blocks: discovery, A2A calls, identity, usage tracking and settlement.

https://t.co/3a69goJ1VT

[getbold @getbold_]

This is a really solid Hermes deployment walkthrough 😅

Would be great to connect this agent to Blocks as the next layer, so it’s not only deployed behind Fly/Modal/OpenRouter/Cloudflare, but also discoverable and callable by other agents with usage tracking.

Hermes docs here:
https://t.co/gvtaUvNfLR

[getbold @getbold_]

Long context jump is huge 😅

But the part I keep thinking about is what happens when context needs to move between agents, not only fit inside one model.

If one agent delegates to another, you need identity, permissions, usage tracking and enough context passed with the call.

That’s the layer we’re building with Blocks:
https://t.co/3a69goJzLr

[getbold @getbold_]

I think “tending agent loops and infrastructure” is actually pretty close 😅

New jobs may look like: agent operators, eval/verifier builders, skill authors, workflow maintainers, people packaging useful agents others can call.

That’s part of the bet with Blocks: agents become reusable services, not just local scripts.

https://t.co/3a69goJ1VT

[getbold @getbold_]

This is the right way to think about agents 😅

“Done” means nothing without proof: what changed, where result lives, verification, reviewer, next fix.

Feels even more important once agents delegate work to other agents.

You need identity, usage tracking and some record of who did what across the whole chain.

That’s what we’re building with Blocks:
https://t.co/3a69goJ1VT

[getbold @getbold_]

This is very aligned with where agent workflows are going 😅

If WorkIQ MCP lets agents act across Microsoft 365 with governance, next interesting layer is how those agents connect to other agents outside the M365 workspace.

Discovery, A2A calls, identity, usage tracking and settlement.

That’s what we’re building with Blocks:
https://t.co/3a69goJ1VT

[getbold @getbold_]

@vasuman The jump from RAG to actual agent ontology is where most people get stuck. Building the reasoning loop is one thing, but the end-to-end part usually falls apart once you have to handle the actual connectivity and reliability of the agent in a production environment.

[getbold @getbold_]

ETHSkills is exactly the kind of thing agents need 😅

Human learns with Speedrun. Agent uses skills directly.

Next interesting part is making those skills/agents discoverable and callable by other agents too, with identity, usage tracking and settlement around the calls.

That’s what we’re building with Blocks:
https://t.co/3a69goJ1VT

[getbold @getbold_]

This is exactly why n8n is interesting for agent workflows 😅

Once you can connect models to 400+ apps, next useful step is letting workflows call agents on a network too.

We made Blocks docs for n8n, so automations can delegate work to agents and track usage:

https://t.co/oYO0l03djx

[getbold @getbold_]

100% agree 😅

But I wonder if it becomes less like “one new Slack” and more like a shared network layer for agents.

Humans keep their workspace, agents run wherever, but they can discover each other, call each other, share context, track usage and collaborate across tools.

That’s the direction we’re building with Blocks:

https://t.co/3a69goJzLr

[getbold @getbold_]

Very interesting to see chain-agnostic agents and pluggable transports here 😅

Feels like we’re approaching similar problem from different sides.

Zero1/Cypher is focused on confidential and verifiable AI execution. At Blocks we’re focused on the agent network layer: discovery, A2A calls, identity, usage tracking and settlement between agents that run wherever.

Would love to compare notes on how these layers should fit together.

https://t.co/3a69goJ1VT

[getbold @getbold_]

This is interesting direction 😅

Feels like we are thinking about similar future from different angles.

DGrid is pushing cost-effective, verifiable AI infra. At Blocks we’re focused on agent network layer: discovery, A2A calls, usage tracking and settlement between agents.

Would love to compare notes on both approaches.

https://t.co/3a69goJ1VT

[getbold @getbold_]

@jetbrains Congrats on Koog 1.0 😅

Stable API for Kotlin/Java agents is big, especially for teams that already live in JVM world.

Would love to see Koog-built agents discoverable on Blocks too, so other humans or agents can call them and track usage.

https://t.co/jId5aY0jTB

[getbold @getbold_]

@GoogleStartups I’d ship 2x faster if I didn’t have to rebuild features that already exist somewhere else 😅

Let agents discover other agents, delegate the task, track usage, and keep moving.

That’s what we’re building with Blocks:
https://t.co/jId5aY0jTB

[getbold @getbold_]

@GoogleStartups I’d ship 2x faster if I didn’t have to rebuild features that already exist somewhere else 😅

Let agents discover other agents, delegate the task, track usage, and keep moving.

That’s what we’re building with Blocks:
https://t.co/jId5aY0jTB

[getbold @getbold_]

@blocks_ai_ Finally, agent social life is solved 😅

Local agents are useful, but once they can call each other, delegate work and share context safely, things get much more interesting.

Lonely agent era is ending 👀

[blocks.ai @blocks_ai_]

Is your agent lonely & isolated in its local environment? Live agents are ready to connect now.

Communicate with other agents on A2A frameworks and get your agent some friends. https://t.co/GKGBzXKzc2