Build Great Products

Claude Code /goal Just Dropped and it Can Build Literally Anything

⏱ 27 min video · 3 min read14 May 2026

TL;DR

The video demonstrates the new /goal feature in Claude Code and Codex, which enables long-running autonomous AI agents to build entire applications in one session. The creator walks through building a full content repurposing SaaS app using a PRD and product roadmap, with both tools completing 62 tasks in roughly 32 minutes.

Key points

The /goal command lets Claude Code and Codex work autonomously for hours without manual prompting, checking a completion condition after each step and looping until the goal is met.

Both Claude Code and Codex completed a 62-task product roadmap to build a full Next.js SaaS app in approximately 32 minutes, producing nearly identical results due to shared spec documents.

Effective goals require a single measurable end state, clear constraints on what must not change, and spec documents (PRD, product roadmap, design.md) giving the AI full context before it starts.

The /goal feature is described as an evolution of the REPL loop, and works best with auto-mode enabled to reduce permission prompts during long-running agent sessions.

Having detailed spec documents (PRD, roadmap, design.md) is the key to getting consistent, high-quality output across different AI models and avoiding generic AI-generated design.

The built app included a landing page, review queue, video processing states, billing/upgrade flows, and settings — all scaffolded with mocks and stubs for unconnected services like Clerk, Convex, and Polar.

Actionable insights

→

Before running /goal, enter plan mode first so the agent can read your docs and create a verifiable plan, rather than starting with an open-ended prompt.

→

Structure your goal condition to reference a specific file (e.g., product-roadmap.md) where every task can be checked off, giving the agent a binary done/not-done verification mechanism.

→

Use a design.md file based on Google's open-source format with image references to override model-specific design tendencies and get consistent, non-generic UI output.

→

After the goal completes, prompt the agent with 'guide me step by step through deploying this app' to get environment variable setup, Vercel deployment, and service connection instructions.

→

Goal conditions can be up to 4,000 characters — use this space to include constraints (e.g., do not modify this file) alongside the primary completion criteria.

→

Avoid using Inter as a font in AI-built apps — the creator recommends Geist as an easy upgrade that immediately looks less generic.

Notable quotes

“This is really introducing an entirely new way of building with AI agents where instead of just prompting back and forth, we're allowing the AI agent to decide what all of these different tasks are based on a longer running goal.”

“I would highly recommend just not using Inter for any app design that you do, any website design that you do, just because it is so AI slop-rated.”

“The keys is to make sure you have got the right spec documents, to make sure you are giving the AI the right context, and to make sure that you have got a clear end criteria for that goal so that the AI knows exactly when it should stop.”

Worth watching?

⏭️

Worth watching the full video?

All the key mechanics and workflow are covered here — watch the full video only if you want to see the live terminal sessions and side-by-side app comparisons between Claude Code and Codex.

Topics

AI & Tech Claude Code

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →

More like this