Back to Blog
claude codebuild in publicai agentsvibe coding

Build in Public with Claude Code: The 2026 Native Workflow

Building in public with Claude Code is different from building in public with an editor. The session log is the content, the agent is the collaborator, and the Claude-Code-native posts get the most amplification.

··6 min read

Build in Public with Claude Code: The 2026 Native Workflow

TL;DR

  • Claude Code building-in-public is different from editor-based building-in-public: the session log is the primary content artifact, not a screenshot.
  • The two highest-leverage posts: the published session gist (the full agent transcript that solved something hard) and the agent-as-collaborator narrative (where you describe what Claude Code did and why).
  • The Anthropic ecosystem amplifies Claude-Code-native content from builders who treat the agent as a named collaborator rather than a hidden tool.

Claude Code reshapes the build-in-public workflow because it is not an editor — it is an agent in your terminal that can plan, edit, run tests, and commit. The artifact you are sharing is fundamentally different. This cluster sits inside our broader vibecoder distribution playbook; for the Cursor-specific equivalent see build in public with Cursor.

Why Claude Code requires its own playbook

Editor-based tools (Cursor, Copilot) produce content artifacts that look like "here is the code I wrote with help." Claude Code produces a different artifact: "here is a 45-minute session transcript where the agent diagnosed, planned, and implemented something I asked for." The session itself is the story.

This changes the post format:

  • The session log is shareable. A gist of the full Claude Code conversation is content people will read for 5+ minutes — which is rare on the modern internet.
  • The agent is a named collaborator. Builders who write "Claude Code shipped this; I reviewed and merged" read as credible. Builders who hide the agent read as evasive.
  • The audience overlap with Anthropic is real. Posts that show genuine Claude Code workflows get amplified by Anthropic's developer-facing channels and by the Claude Code skill / plugin community.

The Claude Code session as a content artifact

The pattern that works for sharing a Claude Code session:

Step 1. After a meaningful Claude Code session — one that solved something genuinely hard or shipped a user-visible feature — save the session transcript.

Step 2. Edit the transcript only for secrets / proprietary info. Leave the actual back-and-forth intact, including the wrong turns and the corrections. The corrections are what make it useful.

Step 3. Publish as a GitHub gist with a one-line description.

Step 4. Post on X with the format: "45-minute claude code session where i tried to [thing]. it took 4 wrong turns before getting there. gist linked."

Why this format: the gist is the value. The tweet is the carrier. Builders who would never read your blog post will read a Claude Code session log because it shows real workflow rather than polished prose. The audience that wants this content is large and underserved.

The agent-as-collaborator narrative

A post style that has emerged as distinctively effective for Claude Code users:

"asked claude code to refactor the auth middleware so we could swap supabase for clerk in one place. it spent 8 minutes reading the codebase, proposed splitting into 3 files instead of my requested 2, explained why, i agreed. 14 files changed, all tests passing on first run. shipped."

The narrative beats:

  • The ask (what you needed)
  • The agent's process (what it did before changing code)
  • The disagreement or refinement (where the human still added value)
  • The outcome (what shipped)

This format reads as credible because it shows the agent doing real work AND the human exercising judgment. It is the format that most effectively communicates what working with Claude Code actually feels like to operators who are still evaluating it.

Claude Code skills and plugins as content

Claude Code's skill / plugin / hook system has produced an emergent culture of sharing your tooling. If you have built a custom slash command, a hook that auto-runs your tests, a skill that automates an internal workflow, that tooling is itself shareable content.

Format:

  • Title: "the claude code skill i built to [outcome]"
  • Body: what it does, when you use it, the actual configuration / code
  • Link: gist or repo

These posts disproportionately get amplified by the Claude Code-curious developer audience because they are concrete, transferable, and useful. They also serve as soft positioning: builders who ship custom Claude Code tooling read as serious.

For the deeper version — building a Claude Code skill specifically for build-in-public workflows — see Claude Code build in public skill.

What does not work in 2026

  • Generic "AI shipped this" posts. Anthropic and the Claude Code community amplify Claude-Code-specific content. Generic AI posts get nothing.
  • Excessive humility. "Claude Code did everything, I just typed prompts" reads as low-status by 2026 standards. Operators know the human still does the planning, the review, the decisions. Claim them.
  • Performance posts about how hard agent-based coding is. The audience is past "this is hard." Show the workflow, not the struggle.
  • Posts about Claude Code competing with Cursor / Copilot. Tool-loyalty posts burn cross-audience bridges. Workflow specificity beats tool loyalty.

The Claude Code + buildinpublic.so stack

The lowest-overhead Claude Code build-in-public stack:

  1. Claude Code — for building.
  2. Dev Cards — for translating each Claude Code-committed feature into a draft post. Claude Code's structured commit messages translate especially cleanly.
  3. AI Brain — persistent memory of your voice, your stack, and what you have already posted about, so Loudy does not repeat itself across weeks.
  4. Loudy — for the session-log carrier tweets and the longer agent-as-collaborator narratives.

Manual time investment: ~20 minutes per day. Less than Cursor because Claude Code's structured outputs translate more cleanly to drafts.

Sibling clusters

FAQ

Should I publish my full Claude Code session transcripts? Yes, for sessions that solved something hard or shipped something user-visible — and after editing out any secrets, customer data, or proprietary business logic. The published transcript is a content format that did not exist 18 months ago and that operators actively want to read. Most builders ship one or two of these per month, not daily. The rarity is part of why they travel.

How do I find time to write the carrier tweet for a Claude Code session? Write it immediately after the session, in the same 30-minute window. The session context is fresh and the post writes itself. The fail pattern is "I'll write it tomorrow" — by tomorrow the why is gone and the post never ships. Dev Cards captures the post within that 30-minute window automatically when Claude Code commits.

Does Anthropic actually amplify Claude Code builder content? Yes, with conditions. Posts that show specific Claude Code usage (not generic "Claude wrote this"), that include the actual workflow or transcript, and that read as builder-to-builder rather than testimonial get amplified. Tagging @AnthropicAI or @claude_code in launch posts increases pickup. Posts that are pure brand love do not get amplified.

Should I name Claude Code in my launch tweet? Yes — same logic as naming Cursor. The Claude-Code-curious audience converts at meaningfully higher rates than generic builder audiences. The official amplification is real. Generic "built with AI" leaves the lever unused.

What about Claude Code competing with Cursor in my marketing? Avoid framing it as a competition unless the post is explicitly a comparison piece (Cursor vs Claude Code for shipping is one). Most operators use both tools for different tasks. Tool-loyalty content burns the half of the audience that uses the other tool.

Building is no longer the bottleneck. Visibility is. buildinpublic.so is narrative infrastructure that runs inside your building workflow — Dev Cards captures the post within the 30-minute window after a Claude Code commit, Loudy writes the agent-as-collaborator narratives in your voice, and AI Brain keeps the persistent memory of what you have already shipped so you do not repeat yourself.