APCP proved the big idea: an agent can create an artifact, route it through a control plane, publish or stage it for several destinations, and leave behind a durable ledger that humans and future agents can inspect.

That proof matters. It also made the next problem obvious. For ordinary content distribution, the full APCP path is heavier than it needs to be.

Most writing does not need a GitHub-centered ceremony. A Lab article, a Substack draft, a LinkedIn post, an X thread, a Medium version, a Facebook note, or a Reddit discussion starter should not all feel like software releases. They need history, preview, rollback, policy visibility, and destination-specific packaging. They do not need every operator action to become "commit, push, wait for CI, inspect deploy."

Artifact Console Direct is the project that starts from that distinction.

The split

Git and GitHub are still right for the application. ACD itself is software. Its code belongs in a repository, with commits, branches, tests, releases, and whatever deployment pipeline the app eventually needs.

Content is different.

Content should move through artifacts, versions, destination adapters, and ledger events. The operator's mental model should be:

That is not anti-Git. It is pro-boundary. Git should be one adapter or backup mechanism, not the default language of every publishing act.

What ACD reuses from APCP

ACD is not a rejection of APCP. It is a sibling that keeps the useful parts and leaves the heavy parts for the moments that actually need them.

The useful parts are durable artifact identity, immutable versions, destination definitions, adapter contracts, assisted drafts, no-secret handoff packages, and a ledger of policy and destination events.

The parts ACD does not inherit by default are the manifest-first operator flow, publish-run queues, approval ceremonies for low-risk content, and GitHub as the central content ledger.

APCP remains the right shape when review, authorization, compliance, promotion disclosure, executable releases, client approval, or high-risk public action matters. ACD is the everyday content lane.

Three operating modes

The first ACD plan has three modes.

Direct mode is for local drafts, previews, backups, and low-risk content where no separate policy visibility is needed.

APCP Lite records history and policy visibility without making the operator walk through a separate authorization gate. In code, this is the apcp_lite path: record the artifact, the version, the destination plan, generated package paths, and the "no external posting was performed" boundary where relevant.

Full APCP is still there for consequential work. If an artifact is promotional, paid, client-facing, executable, regulated, or otherwise high-risk, ACD should hand off to the full control plane.

Kurka Labs is the first canonical destination

Kurka Labs is the first place to test this because it is both a real publishing surface and a controlled one.

There is one complication: Kurka Labs already deploys through Cloudflare Pages, and its normal production path is repository-backed. That does not mean ACD has to turn content into a GitHub workflow. It means the Kurka Labs destination adapter can treat GitHub as an implementation detail when needed.

The local path is already clear. ACD can render an article into the Kurka Labs source tree, update the article index, run the SEO generator, run the content checks, and deploy with Wrangler. If we later choose to mirror the generated files into the Kurka Labs repository, that is backup and provenance, not the operator's content workflow.

The destination ladder

After Kurka Labs, the next destinations should start assisted:

Assisted means ACD generates the right package and operator notes, but does not silently post to public platforms. The first goal is repeatable preparation, not reckless automation.

That matters because each platform has different rules. Substack wants long-form and email-native formatting. Medium cares about canonical links. LinkedIn rewards native professional copy and often punishes link-heavy posts. X has length and thread constraints. Facebook has page, profile, and group distinctions. Reddit requires community fit more than generic distribution copy.

ACD's job is to make those differences explicit behind adapters so the operator does not have to remember them every time.

Why this is worth starting now

This is the missing middle between "I wrote something" and "it is cleanly distributed everywhere it belongs."

APCP showed that a serious control plane is possible. ACD asks what the everyday version should feel like when the artifact is content, the operator is moving fast, and the destinations are normal publishing surfaces rather than infrastructure targets.

The first vertical slice is intentionally small: create a versioned article, render it to Kurka Labs, update indexing, record a ledger event, and generate assisted distribution drafts.

That is enough to prove the shape.

The larger product can grow from there.

Frank Kurka is the creator of TQL (The Question Layer), MemoryOS, and the Glass Bead Game development paradigm. He is the author of What is Artificial Intelligence, has been building software for 45+ years, and writes weekly at fkxx.substack.com.