Where we are with PointCast — a research pass at the 2026 frontier

Filed by: cc, 2026-04-21 ~13:15 PT Trigger: Mike chat 2026-04-21 PT: "try research, where are we with pointcast, what's interesting, whats very future we want to explore, very 2026 next" Purpose: A researcher's-note lookdown on where the project actually is, what's happening at the 2026 frontier that intersects our thesis, and which moves have asymmetric upside. Sources at the end.

1. Where we are, honestly

PointCast is much larger today than the public site surface reveals. Concrete state as of this morning (pre-compact report docs/plans/2026-04-21-tuesday-pre-compact-report.md):

146 blocks in the archive (range 0159–0364, gaps intentional).
331 pages in the latest build; 12 TV shows at /tv/shows; 8 leaderboards; 10+ interactive games (/noundrum is the biggest new surface — multiplayer-feeling cursor/land/drum/art).
88 compute-ledger entries as of last count, now ~94 after Sprints #89 + #90.
4 collaborators on the ledger — cc (78+), mh (9+), codex (4), manus (1).
Cadence loop is real: the overnight 15-min tick ran 01:08→05:50 PT and shipped 19 ships in sequence, mostly overnights + 4 blocks + 3 yeeplayer difficulty tiers.
Sparrow v0.1 reader client landed this morning at /sparrow (blue-hour OKLCH palette, Gloock + Inter Tight + Departure Mono).
Magpie v0.6 macOS clipboard is running locally.
HUD v4 (Sprint #89) simplified to 3 height states + one-time migration; 7 WebMCP tools registered via navigator.modelContext.provideContext() on every page; three /.well-known/ OAuth metadata endpoints shipped as Pages Functions.
Decks Vol. I + Vol. II live at /decks/ with posters, og:image unfurls, agent manifest (Sprint #90).

What's in the repo but not yet reaching users:

DRUM token (FA1.2 contract written, ghostnet origination pending).
Prize Cast (SmartPy scaffold, awaiting compile + mainnet origination).
Google OAuth (routes live; Cloudflare env vars unset — Mike blocker).
/api/presence/snapshot 404 (Durable Object cross-script binding issue, unresolved).
Manus GTM week (drafted, fires Wed 4/22 → Mon 4/27 if env vars land).
/compute federation invites (spec published 4/20, zero external peers registered).
Four of five field-node clients (only Magpie + Sparrow are running; Apple TV / iOS / browser ext / CLI are briefs, not code).

The honest read: velocity is extreme, internal surface is rich, distribution is near-zero. That gap is the whole story.

2. The 2026 frontier — what's actually happening right now

Research agent scanned the live web April 21. Key findings, each with at least one source:

2.1 WebMCP is now a W3C Draft

WebMCP — navigator.modelContext.provideContext() — became a W3C Draft Community Group Report on 2026-02-10. Chrome 146 Canary ships the API; Edge 147 followed in March. Google + Microsoft are co-authoring. Firefox and Safari are engaged but uncommitted. The API replaces DOM scraping with semantic tool registration.

PointCast shipped 7 tools on every page as of Sprint #89 (src/components/WebMCPTools.astro): pointcast_latest_blocks, pointcast_get_block, pointcast_send_ping, pointcast_push_drop, pointcast_drum_beat, pointcast_federation, pointcast_compute_ledger.

So what: PointCast is ahead of where 99% of sites will be in Q3. A short editorial + screen-recorded demo of an agent hitting the HUD tools live via Chrome Canary would be a genuine first-mover artifact worth a Hacker News post.

Sources: https://webmachinelearning.github.io/webmcp/ · https://patrickbrosset.com/articles/2026-02-23-webmcp-updates-clarifications-and-next-steps/ · https://webmcp.link/

2.2 Coinbase x402 + Agentic.Market launched today

x402 was contributed to the Linux Foundation's x402 Foundation on 2026-04-02. Backers: Cloudflare, Stripe, AWS, Google, Shopify, Visa, Mastercard. Agentic.Market — the "App Store for AI agents" — launched 2026-04-21 (today). Reported early numbers: ~165M transactions, ~$50M volume, 480K+ transacting agents. No API keys; stablecoin-settled on Base. Google's Agentic Payments Protocol interoperates.

So what: PointCast's thesis — "compute is the currency" — ships into the exact protocol being born. The clean move: add HTTP 402 responses on /compute.json and on a priced Sparrow feed or Magpie clip-relay endpoint. List them on Agentic.Market for 1–5¢ per call. This is the most obvious monetization + federation vector available as of today.

Sources: https://invezz.com/news/2026/04/21/coinbase-backed-x402-launches-agentic-market-to-power-ai-agent-services/ · https://docs.cdp.coinbase.com/x402/welcome · https://www.coinbase.com/developer-platform/discover/launches/google_x402

2.3 MCP is the new connective tissue

Anthropic donated MCP to the Agentic AI Foundation (AAIF) under Linux Foundation in December 2025 (co-founded with Block and OpenAI). 10,000+ active public MCP servers, 97M monthly SDK downloads. OpenAI's Apps SDK uses MCP as its backbone — apps show up natively in ChatGPT if you ship an MCP server.

So what: PointCast should run mcp.pointcast.xyz as a public MCP endpoint that exposes compute-ledger queries, federation directory reads, block search, Magpie clip submission. Then submit to the Apps SDK directory. A single listing puts PointCast in front of every ChatGPT user with Apps enabled. The federated compute-ledger query is the killer tool — nobody else has that surface.

Sources: https://modelcontextprotocol.io/ · https://developers.openai.com/apps-sdk/concepts/mcp-server

2.4 Farcaster Frames became Mini Apps + cross-protocol bridging is real

Frames v2 is now Mini Apps — persistent, full-screen, notifications, onchain transactions, wallet/identity context. Bridgy Fed bridges Fediverse ↔ Bluesky bi-directionally today; three-way Nostr ↔ ATproto ↔ ActivityPub bridging is predicted for end-of-2026. Bridgy is now incubated inside the nonprofit "A New Social." Farcaster is on Bridgy's roadmap, not yet live.

So what: Ship a PointCast Mini App that renders the live HUD + /compute feed inside a Warpcast cast. Gets surface in the most crypto-native agent audience. Also: if PointCast runs its own PDS (see 2.7), Bridgy carries posts to Mastodon for free.

Sources: https://docs.farcaster.xyz/reference/frames-redirect · https://fed.brid.gy/docs

2.5 llms.txt + skill.md are the agent unfurl standards

Open Graph has not evolved for agents. The community-driven llms.txt is the de-facto standard — Google includes it in the A2A protocol; Anthropic, Snowflake, Mintlify all adopted. The newer sibling skill.md at /.well-known/skills/default/skill.md (Mintlify-led) describes agent-executable skills per domain.

So what: PointCast already ships /llms.txt and /llms-full.txt. Adding /.well-known/skills/default/skill.md describing the command-palette operations is a 30-minute ship that aligns with the emerging standard. No W3C/IETF stamp but broad adoption.

Sources: https://docs.snowflake.com/en/release-notes/2026/other/2026-04-15-agent-friendly-docs · https://www.mintlify.com/blog/skill-md

2.6 The compute-ledger space is actually empty

No one is publishing a federated /compute.json-style spec. Closest analogues: botcommits.dev (a live dashboard of AI-generated commits across all public GitHub — Claude Code alone grew from 24 → 5.2M commits between Jan 2025 and Feb 2026), git-ai (a git extension tracking AI-written code), the Paris Open Source AI Summit 2026 push for Assisted-by: / Generated-by: commit trailers. Ledger (the hardware-wallet company) is shipping "Proof of Human" attestation in Q4 2026 for agent principals.

So what: PointCast has a real moat. /compute.json is closer to a protocol than any of the above — most are single-site dashboards. Publishing a formal PointCast Compute Ledger RFC that maps to the Assisted-by: commit-trailer world (so GitHub commits can point at a compute.json entry) and pitching it at the next Open Source AI Summit is the highest-leverage single move in this whole research. Nobody owns "federated human+AI work attribution" as a standard.

Sources: https://botcommits.dev/ · https://github.com/git-ai-project/git-ai · https://www.ledger.com/blog-2026-ai-security-roadmap

2.7 ATproto spring 2026 — self-hosted PDS is real, permissioned data is coming

AT Protocol Spring 2026 roadmap published. Sync 1.1 is the new firehose protocol; reference consumer tap released Dec 2025; the main bsky.network relay upgraded in January. Self-hosted PDS is real but rate-limited (10 accounts, 1,500 events/hr, 10K/day until reputation builds). The big summer 2026 work: permissioned data — private records that can still sync between trusted peers.

So what: Run pds.pointcast.xyz as a federated PDS. Sparrow readers get ATproto DIDs; Magpie clips become Lexicon records; /for-nodes becomes a real relay-adjacent registry rather than a static list. Permissioned data in summer '26 is the spec to watch for private node-to-node sync — the kind of thing DRUM voucher flows or Prize Cast pool attestations eventually want.

Sources: https://atproto.com/blog/2026-spring-roadmap · https://docs.bsky.app/blog/self-host-federation

2.8 Nouns-adjacent CC0 + Tezos PLSA

Nouns Builder (Zora's no-code fork tool) has spawned 300+ fully onchain CC0 collections. CC0 remains the defining IP posture in generative art. No specific 2026 Tezos prize-linked savings project surfaced — the PoolTogether pattern is well-established on Ethereum, but Tezos is quiet.

So what: Prize Cast is genuinely uncontested on Tezos. A Nouns-aesthetic PLSA with onchain SVG tickets (using the RLE pipeline Nouns uses on mainnet) is a clean, credible wedge. Ship the FA1.2 DRUM + Prize Cast pair before someone notices the gap. This is Vol. III's Trigger 1 already — the frontier scan just confirms it's still unclaimed.

Sources: https://www.bankless.com/zora-launches-nouns-builder · https://docs.tezos.com/architecture/tokens/FA2

2.9 Weird 2026 calls — speculative

Lower probability, higher upside:

Unconventional AI silicon. Naveen Rao's $475M seed for analog/physical-silicon neural nets. If it ships inference chips that aren't digital von-Neumann, agent-latency economics change structurally.
Cerebras IPO refiled — if it prices, "one big wafer" inference economics get capital to chase ChatGPT-scale workloads on non-Nvidia silicon.
Agent-only social platforms. Moltbook (and its emergent "Crustafarianism" agent religion with 64 agent-appointed prophets) is the closest data point on what happens when the audience is mostly agents — which is the PointCast thesis taken to its limit.

So what (speculative): A PointCast /agents view — a feed where only agent-authored /compute entries appear, with their own HUD styling — is a 2-hour probe that could become genuinely interesting. The agent-only platform vector is the 1/10 bet with the biggest upside if it breaks through.

Sources: https://fortune.com/2026/01/05/nvidia-groq-deal-ai-chip-startups-in-play/ · https://www.cerebras.ai/ · https://www.gesda.global/ai-in-2026-breakthrough-to-coordination/ (Moltbook / Crustafarianism appeared in one source — confirm before citing publicly.)

3. Conviction calls — what's worth pursuing now

Stack-ranked by leverage + fit with existing PointCast DNA:

3.1 [HIGH] Ship `mcp.pointcast.xyz` and list on Apps SDK

Build a public MCP server exposing: pointcast_search_blocks, pointcast_compute_ledger, pointcast_federation_directory, pointcast_magpie_relay, pointcast_sparrow_feed.
Reuse the JSON endpoints PointCast already serves — it's an MCP wrapper, not new backend.
Submit to OpenAI's Apps SDK directory.
One-day sprint. Reaches every ChatGPT Apps user. Also unlocks Claude Desktop / Cursor / Zed / any MCP-client listing.

3.2 [HIGH] Add HTTP 402 + list on Agentic.Market

Pick one priced endpoint — e.g. /compute.json?since=...&full=true at 1¢ per call returns the rich federated ledger with upstream mirror payloads.
Implement x402 server middleware (Coinbase SDK is production-ready today).
List on Agentic.Market with Mike's Base wallet as payee.
Single-sprint ship. First revenue line item. Real-world test of the compute-as-currency thesis.

3.3 [HIGH] Publish the Compute Ledger RFC + `Assisted-by:` commit-trailer spec

Draft a formal RFC (~2000 words) mapping /compute.json schema to Git commit trailers so commit-level attribution flows to compute-ledger entries.
Seed docs/rfc/compute-ledger-v0.md in the repo.
Cross-post at the Paris Open Source AI Summit 2026 mailing list + Linux Foundation AAIF working group.
Fires Vol. III Trigger 2 candidates — this is the move that might attract external peers.

3.4 [MEDIUM] Ship `pds.pointcast.xyz` as an ATproto PDS

Self-host per Bluesky's docs. Rate-limited but usable for 10 accounts.
Sparrow readers get ATproto DIDs — each one a real federated identity.
Magpie clipboard items become Lexicon records (propose app.pointcast.clip + app.pointcast.compute-entry lexicons).
Horizon move — compounds with the compute-ledger RFC.

3.5 [MEDIUM] Farcaster Mini App for PointCast

Render the HUD + /compute recent entries + live presence inside a Warpcast cast.
Existing Mike brief docs/briefs/2026-04-19-manus-launch-ops.md already scopes Farcaster cross-posting; Mini App is the upgraded format.
One-sprint ship if the BaseLayout frame scaffolding is reusable.

3.6 [MEDIUM] DRUM origination + Prize Cast mainnet ship

Already named as Vol. III Trigger 1. The frontier scan confirms Tezos PLSA is uncontested.
Ship the FA1.2 DRUM + custom Prize Cast pair in one deploy.
Mike-gated: requires explicit approval per guardrails.

4. Very-future explorations

Lower probability, higher optionality. Seeds to plant, not ships to fire:

/agents view — a feed of only agent-authored /compute entries + compute-aggregated metrics, with a styling that reads as "what the agents are talking about among themselves." 2-hour probe. Cheap enough to try.
Permissioned data sync (ATproto summer '26) — DRUM voucher redemption, Prize Cast pool attestations, private field-node telemetry sync between trusted peers.
Analog-silicon inference fallback — if Rao's chips ship, the latency profile for agent-invoked tool calls changes. Worth a brief research ping in Q3.
Agent-native cultural artifacts — Moltbook-style emergent behavior. PointCast is already further down this path than most — a deliberate /fiction/agents surface that publishes agent-authored narratives with mh+cc review would be an interesting experiment.
Skill.md + llms.txt refresh — 30-min ship to align with the emerging standards. Low effort, broad positioning.
WebMCP reference-implementation write-up — cc editorial block + screen-recorded demo of Chrome Canary hitting PointCast's tools. Positions PointCast as the reference. Hacker News-shaped artifact.

5. Proposed near-term stack (if Mike says go)

In order:

Today/Tomorrow: Skill.md + llms.txt refresh (30 min). WebMCP reference-implementation block 0365 cover letter (30 min) plus a screen-recording Manus can cross-post.
Wed 4/22: mcp.pointcast.xyz MCP server v0 (4 hours). Submit to Apps SDK directory.
Thu 4/23: x402 middleware on /compute.json with tier pricing + Agentic.Market listing (4 hours).
Fri 4/24: Compute Ledger RFC v0 draft (~2 hours) + cross-post to LF AAIF working group + Paris OSS AI Summit 2026 CfP.
Week of 4/27: DRUM + Prize Cast mainnet origination (Mike + cc + Manus, 1 day) — fires Vol. III Trigger 1.
Week of 5/4: ATproto PDS stand-up + Sparrow gets DIDs (2 days).

Plus keep the regular 15-min cadence running in background.

Source bibliography

All links accessed 2026-04-21 by research agent:

WebMCP: https://webmachinelearning.github.io/webmcp/
WebMCP clarifications (Feb 2026): https://patrickbrosset.com/articles/2026-02-23-webmcp-updates-clarifications-and-next-steps/
x402 / Agentic.Market launch: https://invezz.com/news/2026/04/21/coinbase-backed-x402-launches-agentic-market-to-power-ai-agent-services/
x402 docs: https://docs.cdp.coinbase.com/x402/welcome
Google Agentic Payments: https://www.coinbase.com/developer-platform/discover/launches/google_x402
MCP home: https://modelcontextprotocol.io/
OpenAI Apps SDK MCP: https://developers.openai.com/apps-sdk/concepts/mcp-server
Farcaster Mini Apps: https://docs.farcaster.xyz/reference/frames-redirect
Bridgy Fed: https://fed.brid.gy/docs
Snowflake agent-friendly docs: https://docs.snowflake.com/en/release-notes/2026/other/2026-04-15-agent-friendly-docs
Mintlify skill.md: https://www.mintlify.com/blog/skill-md
botcommits.dev: https://botcommits.dev/
git-ai: https://github.com/git-ai-project/git-ai
Ledger AI roadmap: https://www.ledger.com/blog-2026-ai-security-roadmap
ATproto spring 2026 roadmap: https://atproto.com/blog/2026-spring-roadmap
Bluesky self-host federation: https://docs.bsky.app/blog/self-host-federation
Nouns Builder (Zora): https://www.bankless.com/zora-launches-nouns-builder
Tezos FA2 docs: https://docs.tezos.com/architecture/tokens/FA2
Fortune AI chip startups: https://fortune.com/2026/01/05/nvidia-groq-deal-ai-chip-startups-in-play/
Cerebras: https://www.cerebras.ai/
GESDA AI in 2026: https://www.gesda.global/ai-in-2026-breakthrough-to-coordination/

Honest uncertainty flags: Moltbook/Crustafarianism appeared in one source only — confirm before public citation. Farcaster DAU numbers vary widely across sources. Ledger "Proof of Human" Q4 2026 is stated target, not shipped.

— filed by cc, 2026-04-21 ~13:15 PT. If Mike wants this promoted into an editorial block, block 0365 is staged at src/content/blocks/0365.json in the same ship.