cd /news/ai-agents/show-hn-planwright-a-control-plane-f… · home topics ai-agents article
[ARTICLE · art-57890] src=planwright.tools ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Show HN: PlanWright – A control plane for AI coding agents

PlanWright launches as a control plane for AI coding agents, inverting planning and acceptance ceremonies to match agent speed. The tool synthesizes raw inputs like transcripts and Slack into structured objectives, then triages acceptance by routing only judgment-critical items to humans, with every ruling cryptographically signed for auditability.

read5 min views1 publishedJul 13, 2026
Show HN: PlanWright – A control plane for AI coding agents
Image: source

The Control Plane for Agent Labor

Speed moved to the middle and left the ceremonies behind. Planwright inverts the two that are still human-paced: it turns your chaos — transcripts, decks, email, Slack — into objectives, and triages acceptance so you rule only on what needs judgment. Planning and acceptance finally move at agent speed, and every ruling is signed.

Claude Desktop · MCP

↳ “SOC 2 export bundle” · scheduled

The Problem

You bought the agents. #

Where did the speed go?

Coding used to be the bottleneck. Not anymore. Now the constraint is everything you didn't automate:

Planning

A human hand-typing tasks at UI speed to feed machines running thousands of times faster.

Acceptance

A senior engineer reading pull requests one at a time — which is exactly where velocity goes to die.

Make the middle free and infinitely fast, and the edges become the ceiling. That's not a tooling gap. It's Amdahl's law.

The Fix

Invert the two ceremonies coding agents can't remove. #

Invert planning: chaos in, objectives out.

Your planning input was never tidy — it's meeting transcripts, decks, email threads, Slack. From inside Claude Desktop, Planwright synthesizes that raw signal into structured objectives with machine-checkable acceptance criteria. The old ceremony had a human write the plan and the tool record it. Planwright flips it: the agent drafts, you approve — at the speed of describing what you want.

Invert acceptance: triage, don't drown.

When objectives complete, Planwright picks them up and triages acceptance in Claude Cowork — combing the plain, mechanical criteria and clearing what it safely can, then routing the critical, judgment-bound items to you. The machine doesn't accept; it sorts. You stop reading every diff and start ruling only on what actually needs a human, so your scarce judgment gets concentrated instead of flooded. Every ruling is signed.

The Acceptance Gate

The machine doesn't accept — it sorts. #

The state machine ensures a human rules on every judgment call before work reaches “done.” Mechanical criteria get cleared automatically; critical decisions route to you. Your sign-off is the change-approval control point your SOC 2 auditor is looking for.

Human sets the objective and schedules it for agent pickup.

Agent claims the work, plans, codes, and records diffs via MCP.

Agent requests acceptance. A human reviews the diff, tests, and signs off.

Human approves. The audit record is sealed and hash-chained.

Every lane transition produces a hash-chained, cryptographically signed audit record. The chain is independently verifiable on the public trust page.

How It Works

Three primitives. One throughput unlock. #

Turn chaos into objectives.

In Claude Desktop, Planwright synthesizes your transcripts, decks, email, and Slack into structured objectives with machine-checkable acceptance criteria. You approve intent instead of hand-authoring tasks.

Triage acceptance, don't drown in it.

When objectives complete, Planwright combs the mechanical criteria in Claude Cowork and clears what it safely can, routing only the judgment calls to you. The machine doesn't accept — it sorts.

Every ruling is signed.

Machine-cleared throughput stays accountable: each human decision is a hash-chained record, not an ad-hoc PR read.

The Board

Objectives flow. Agents execute. Humans approve. #

Add webhook retry logic

Migrate auth to PKCE flow

Implement ECDSA P-256 audit signing

Add RLS policies for workspaces

SSE transport for MCP server

Database schema v1

How It Works

Five steps. Zero busywork. #

Write an objective

Describe the outcome you want. Not the steps. Not the implementation. A clear, atomic objective your agent can own end-to-end.

Your agent claims it via MCP

Claude Code connects to your Planwright board, picks up scheduled objectives, and starts working. No copy-pasting tickets. No context-switching.

Agent plans, executes, requests acceptance

The agent decomposes the objective, writes code, runs tests, and moves it to the acceptance lane. Every step recorded in the audit chain.

You review and sign off

Every acceptance is a cryptographic signature. Not a checkbox. A real approval with your identity bound to it.

Every step hash-chained and signed

Immutable audit trail from objective creation to final merge. SOC 2 evidence that writes itself.

Compliance

SOC 2 auditors are raising the bar on AI-generated code. #

Under the 2026 Trust Services Criteria, auditors increasingly expect immutable, tamper-evident audit trails and documented change-approval controls for AI-driven changes. Most PM tools can't produce them. Planwright ships this as a core primitive — hash-chained, ECDSA P-256-signed, from the first objective to the final merge.

Get Started

One command. Every agent. #

No tokens. No config files to hand-edit. Run the install command — your browser opens for GitHub login, and you're connected.

claude mcp add planwright --transport http https://mcp.planwright.tools/mcp

Zero-config. Run this command and your browser opens for GitHub login. The connection is cached for future sessions.

For CI and service accounts that can't open a browser, generate a static token in Settings → MCP Token after signing in. Audit records attribute to the token creator.

Need more detail? Read the full Claude Code setup guide →

Pricing

Start free. Scale when you're ready. #

Every plan includes the MCP server, audit chain, and board. Pay for seats and retention depth.

Free

For solo developers trying agent-native planning.

  • Up to 3 seats
  • 1 project
  • 30-day audit retention
  • Community support

Get started free

Team

For small teams shipping with agents every day.

  • 4–10 seats
  • Unlimited projects
  • 1-year audit retention
  • Public trust page
  • GitHub App integration

Start free trial

Business

For teams that need compliance out of the box.

  • 11–50 seats
  • Unlimited projects
  • 3-year audit retention
  • Custom domain trust page
  • Priority support
  • Advanced analytics

Start free trial

Enterprise

For organizations with regulatory requirements.

  • 51+ seats
  • 7-year audit retention
  • SSO / SAML (roadmap)
  • SOC 2 attestation letter (roadmap)
  • Dedicated support
  • Custom integrations

Contact sales

FAQ

── more in #ai-agents 4 stories · sorted by recency
── more on @planwright 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/show-hn-planwright-a…] indexed:0 read:5min 2026-07-13 ·