A planning tool for AI coding agents from BrainGrid AI, Inc. Its concept is “Plan first. Build second.” — aimed at avoiding the situation where you jump straight to code and then spend your time fixing a broken app. A planning agent asks follow-up questions about a vague one-line prompt and turns it into a spec that includes the goal, the context, and acceptance criteria. Large features are broken down into tasks small enough for autonomous agents to handle, and can then be handed off to coding agents such as Claude Code, Cursor, OpenAI Codex, and Google Gemini via an MCP server, the CLI, or plain copy and paste. The product has added features since 2025, and in March 2026 it announced a $1 million pre-seed round led by Menlo Ventures.
Note that as of August 2026 the official site displays “BrainGrid is now Plansmith” and new signups are closed (existing accounts continue to work). Anyone starting now is directed to the successor product, Plansmith. What follows is based on the information published for BrainGrid.
Key Features
- Turns vague prompts into language through questions: For a short request like “I want to build X,” the planning agent responds with questions about the purpose, assumptions, and constraints. Edge cases and hidden complexity get written down before implementation begins
- Specs with acceptance criteria in minutes: It produces a specification with the goal, relevant context, and acceptance criteria. Because “what counts as done” is settled up front, an agent’s output can be checked mechanically
- Breaks large features into small tasks: Work is split to a granularity an autonomous agent can handle in one pass and converted into structured, implementable tasks — a design meant to reduce the failures that come from throwing a whole feature at an agent
- Integration via MCP server and CLI: Coding agents can read and act on tasks directly through the Model Context Protocol. The CLI includes automatic branch creation (
braingrid requirement create-branch) andbraingrid requirement review, which streams an acceptance review against a pull request - GitHub integration and repository context: It supports connecting repositories to existing projects, code search across GitHub, and branch-aware behavior that respects the selected branch. Codebase summaries are loaded on demand to avoid overflowing the context window
- A loop that covers build and verification: The official site describes a flow where, after the spec is approved, a builder agent implements it, verifies each requirement with attached evidence, and then ships to a live environment
Pricing
| Plan | Monthly | Annual | What’s included |
|---|---|---|---|
| Solo | $42 per user | $500 per user (equivalent to two months free) | 1 user, 10,000 credits per month, build-ready specs, integration with Linear, Jira, Asana, and others |
| Team | $167 per user | $2,000 per user (equivalent to two months free) | Everything in Solo, 50,000 credits per user per month, comments and sharing, single sign-on, audit logs, priority support. Comment-only seats are $5 each |
| Enterprise | Contact sales | Contact sales | Everything in Team, self-hosted GitHub / GitLab / Bitbucket, volume pricing, dedicated onboarding, custom contracts |
There is no free plan.
Pricing is as of August 2026. The official pricing page also carries the notice about the move to Plansmith, so please check the official site for current terms.
Pros & Cons
✅ Pros
- Because acceptance criteria are fixed before implementation, an AI agent’s output can be judged on whether it actually meets them
- Questions surface the ambiguity hidden in a one-line request, making it easier for non-engineers to articulate a specification
- Not tied to a single agent — output can be passed to Claude Code, Cursor, Codex, or Gemini via MCP, the CLI, or copy and paste
- Task granularity is tuned for autonomous agents, reducing the breakdowns caused by handing over an entire feature at once
- Branch creation and pull-request acceptance reviews are available from the CLI, so it fits into an existing GitHub workflow
⚠️ Cons
- New signups are closed, so newcomers are pointed to the successor product, Plansmith
- There is no free plan, and even Solo is $42 per month, which is not cheap for individual use
- The credit-based model makes costs hard to predict in months with a lot of planning cycles
- Adding a planning step is heavy for small fixes or one-shot requests
- The generated spec still needs review; it cannot be trusted wholesale
Comparison with Similar Services
| Criteria | BrainGrid | Kiro | Task Master | Devin |
|---|---|---|---|---|
| Provider | BrainGrid AI, Inc. | Amazon Web Services | Open source community | Cognition |
| Positioning | Focused on planning and spec generation | An integrated environment with spec-driven development built in | A CLI for task breakdown and management | An autonomous agent that also implements |
| Main output | Specs with acceptance criteria, plus small tasks | Specs, design, tasks, implementation | Task lists (JSON / Markdown) | Working code and pull requests |
| Agent integration | Hands off to external agents via MCP and CLI | Self-contained within the environment | Various agents via MCP | Runs on its own agent |
| Intended users | Non-technical planners through developers | Developers | Developers | Development teams |
Who Is It For
- Anyone who keeps asking an AI coding agent for something and receiving a result that misses the intent
- Non-engineers with an idea who want it shaped into a specification they can hand to developers or agents
- Teams that want breaking large features down for agents to be a repeatable process rather than manual work
- Development leads who want acceptance criteria settled first so AI output can be reviewed objectively
Summary
BrainGrid isolates the planning step that AI coding agents struggle with, closing gaps through questions and converting the result into a spec with acceptance criteria. It assumes implementation is left to an agent, so it fits people already using Claude Code or Codex who are frustrated by inconsistent output. That said, new signups closed as of August 2026 and development has shifted to the successor product, Plansmith. Anyone considering it now should check the notice on the official site and weigh the migration path as part of the decision.