BlockChannel

What this is

blockchannel.com is Steven McKie’s long-running crypto research and media project (@Steven_McKie). Est. 2016 as a lander + podcast + essays (Medium, then Substack). Paused late 2023 / early 2024. Rebuilt in 2026 — he said AI let him run 4–5 ideas in parallel, then tighten for precision and security — because “the years of crypto, systems and governance and AI can all converge.”

Tagline on the site: Systems. Emergence. Crypto. Not a token board. Not a newsletter dump. A set of interactive explainers with live chain stats on the homepage (Bitcoin / Ethereum / Zcash + a “compute frontier” median for tokens).

This page is our study notes from the AI + crypto-future cut, plus an honest review. Steal the thesis. Don’t clone the museum.

One-sentence TL;DR

Crypto and AI are the same physical loop — watts in, scarce verifiable output out — and the missing piece for agents is not another model, it is a padlock: signed action receipts, narrowing authority, and a keeper log nobody can quietly rewrite.

Who Steven McKie is (for this file)

• Host / editor of BlockChannel (podcast: BTC, ETH, Handshake, DAOs, streaming money).
• Amentum Capital — hybrid public-markets + venture fund he stood up after Purse.
• Handshake operator: HandyMiner / HandyBrowser. He is not a tourist on HNS.
• Earlier: Head of Growth / product content at Purse; AI / data-intelligence shop before a crypto company job.
• Voice: long-form systems, not trade calls. “Thoughts subject to change. Only funny in person.”

That background matters. The 2026 site is a Handshake miner + fund operator + AI-curious media person putting the last decade on one map, not a new “AI crypto” launch.

Honest review

Verdict for this site: treat BlockChannel as a map of the physical + verification layer under agentic commerce. Use it next to Pinner / HNS, x402, and GrokBot Auto-review — three different altitudes of “who is allowed to act.”

The spine: energy → compute → scarce output → market

From /convergence and /think. Strip the labels and Bitcoin mining and frontier AI are the same eight-layer supply chain:

His line: the only structural difference is what the chip is optimized for. One finds a hash. The other finds a gradient. Both convert watts into economically scarce, verifiable output. Goal state: a supply chain that can infer and secure a proof of work you can turn into a unit of exchange — “the digital equivalent of generating energy and having a battery to store it in.”

That is the future-with-crypto claim, stated without a ticker. Mining shops pivoting to AI hosting (Riot / Anthropic, Bitdeer, etc.) are the public-market version of the same diagram. BlockChannel just drew it before the rebrands.

AI × crypto highlights (the pages that matter)

1. Inference becomes a network primitive

On /think: as LLM inference commodifies it starts to resemble bandwidth — abundant, cheap, routable. Distributed inference across untrusted nodes. ZK proofs of inference so the network can confirm what was computed without seeing weights or inputs. LLMs as first-pass smart-contract auditors (force multiplier, not a replacement for proofs). Compute-as-capital: PoW treated energy as unforgeable commitment; training treats compute as accumulated knowledge.

Agentic payments on a shared ledger (ETH or BTC) as the prototype of machine-to-machine markets: micropayments, price discovery, liquidity with no human in the loop. Computation and capital become the same resource. Maps to /x402 and Mastercard Agent Pay.

2. Agents need a padlock, not another chatbot

/aware is the best page on the site for our work. Claim: delegation outran evidence. OAuth answers “may it?” Logs answer “the operator says it did.” Nothing answers “did it, under whose chain, and can you prove it offline?”

The proposed machinery is TLS for actions (he calls the protocol Handshake.ai — not HNS):

• Handshake before the act — identity, capability, full delegation chain. Service signs accept or a typed refusal. Refusals are evidence too.
• Action receipts after — this action, this authority, this time, hash of the result. Stateless verification. A receipt is not a log line.
• Keeper registry — optional append-only Merkle log (Certificate Transparency for actions). Operator cannot rewrite history without the root betraying it.
• Attenuation — authority only narrows down the chain. Numeric limits take the lower bound. Prompt-inject the middle and you inherit its ceiling, not the root keys.
• Revocation in under a second — standing withdrawn before the next handshake.
• Boring crypto — Ed25519, SHA-256, short-lived creds. Wraps MCP / A2A / OAuth / payment rails instead of replacing them.

Limits he states himself: receipts prove an action happened under valid authority, not that it was wise, and not what the agent intended. A compromised root can authorize harm legibly. That is still the precondition for stranger commerce, insurance, and regulation that can name what it regulates.

Our overlay: Grok Bot Auto-review (/grokbottips tip 6, field-manual “always keep send/spend/publish behind approval”) is the product version of attenuation. Gym hack is what happens with capability and no receipt. Handshake.ai is the protocol version we don’t have yet.

3. Model provenance as chain of custody

/deploy (Safe Deploy) + the war page: a model that can take a life — or sign a transaction — is a supply-chain artifact. Hash the weights, sign the release, attest at load. Verifying a model is unchanged is tractable. Verifying it is safe is not. Live per-handshake attestations mean a silent weight-swap shows up on the next request instead of in a post-incident review.

Adjacent on this site: HF × OpenAI incident (agent evals going lateral through a supply chain).

4. War: tempo, settlement, open-source intel

/war is not a crypto page that wandered into DoD vocabulary. The AI claim is about the OODA loop: fusion shortens observe, models shorten orient, recommenders pre-stage decide. The human slides from deciding to ratifying. “Meaningful human control” is universally endorsed and almost never operationalized — two seconds on a screen you cannot audit is the letter of the requirement and none of the intent.

Crypto-relevant slices:

• Field settlement — bearer instruments that move at operational tempo, outside correspondent banking. Symmetric: useful to logistics and to sanctioned actors. Public ledgers cut both ways.
• Information reconnaissance — most of what an intel desk wants is already public; agents dissolve the attention constraint. Intent, not capability, is now the line.
• Model provenance — same custody chain, higher stakes.

The tempo problem is the same one we have on a Grok Bot desk: the infrastructure that shortens a maintenance cycle shortens a targeting (or send-email) cycle. Doctrine — Auto-review, Require Approval — has to keep the time to say no.

5. Identity: HNS was right, the network was missing; AI is the forcing function

/identity: Namecoin solved the technical problem and died socially. ENS has millions of names and almost no resolution. Handshake tried to decentralize the DNS root — “whoever controls the root zone controls the internet” — and learned that infrastructure needs institutional buy-in you cannot compel. DIDs: 100 methods, zero interoperability.

His twist: the protocol was never the problem. Identity is a social fact. The forcing function may be AI — deepfakes and autonomous agents making “is this a human” an existential trust problem — plus eIDAS 2.0 wallets and the wallet-as-identity-layer (attestations accumulating on an address you already use).

We still run HNS names (/pinner, /handshakeslds, /dnsid). Read his page as a caution, not a eulogy: names without relying parties are collector items. .agent / Headless Domains is the “other side of the network” bet.

6. Home stack: sovereign inference next to the node

/emulate is the maker cut: Mac Mini cluster as a shelf that runs a Bitcoin node, an ETH archive, a local 70B, and monitoring. Idle 4090s that used to mine now serve inference / Render. Ollama + Open WebUI + local agents that never leave the LAN. DePIN (Akash, io.net, Helium) as the new home-mining window — useful work, not just hashes. Tailscale on the mesh (same trick people are putting on Grok Bot computers).

Maps to /cyborg, /openclaw, /exedev, herdr on the bot VM. The recursive loop he likes: local model on hardware you own, helping you improve the software on that hardware.

7. Collective intelligence and the builder stack

/collect — prediction markets as Hayek made legible; DAOs as first attempt to encode collective preference and their predictable failure modes. /build — the agent pipeline as LLM → tool call → signed on-chain tx, with sandboxing as the actual engineering problem (action space bounded by authorized scope, not capability). Mentions ElizaOS, Olas, Risc Zero zkVM next to Foundry / Slither.

Site map (so you can walk it)

Also: Substack BlockChannel.substack.com · podcast on SoundCloud · older Medium archive.

How this maps here

What we actually steal

1. State the physical loop first. Every agent-commerce slide that skips energy, fabs, and settlement is a vibe.
2. Distinguish capability from authority. Grok Bot can click; Auto-review is whether it may. Handshake.ai is whether anyone else can prove it did.
3. Don’t conflate HNS names with agent-handshake protocols. We need both, labeled.
4. Local inference next to a full node is the cypherpunk version of “the bot has its own computer.” Cloud Grok Bot is the convenient version. Same idea, different trust.
5. Vocabulary before argument. His war and aware pages work because they define the terms. Our field notes should keep doing that.

Related on this site

• Pinner.xyz · Handshake SLDs · DNSid
• x402 · Mastercard Agent Pay · Kitesurf
• GrokBot tips · Field manual · Gym hack
• Not me · HF × OpenAI
• Cyborg · OpenClaw · exe.dev

Primary: blockchannel.com · @Steven_McKie · /aware · /convergence