The AI Startup Launch Checklist

We launched DenchClaw on Hacker News. The Show HN hit 147 points, 124 comments, and Garry Tan tweeted about it. We were prepared for most of what happened. We weren't prepared for a few things.

This checklist is based on what we did right, what we wish we'd done, and what we got wrong. It's specific to AI products with a technical audience, but most of it generalizes.

Product Readiness#

Core experience#

• The main value proposition works end-to-end without bugs
• A new user can get to value in under 10 minutes with minimal guidance
• The product handles empty states gracefully (first-time users see something useful, not a blank screen)
• Error messages are helpful, not cryptic stack traces
• The product works on a fresh machine (test on a machine with no dev environment)

Documentation#

• README explains what the product is in the first three sentences
• Installation instructions are complete and tested on macOS + Linux
• At least one tutorial for the most common use case
• FAQ covering the top 5 questions you anticipate

AI-specific requirements#

• What model(s) does your product use? Is this documented?
• What data does the product send to external services? (Especially important for local-first/privacy claims)
• What does the product do when the AI service is down or the user has no API key?
• Are AI-generated outputs clearly labeled as AI-generated?

Technical Readiness#

Infrastructure#

• Product can handle 10x normal traffic without falling over (especially important for HN launches)
• Database backups are configured
• Error logging is set up so you know when things break
• Deployment is automated (not manual SSH)

Security#

• No API keys or secrets hardcoded in source code
• Public-facing APIs have rate limiting
• User data is handled according to your stated privacy policy
• Dependencies are up-to-date (run npm audit or equivalent)

For local-first products (like DenchClaw)#

• The install script works on a clean machine
• The product doesn't phone home with user data unexpectedly
• Uninstall is clean and documented

Legal and Business Readiness#

Basic legal#

• Company is incorporated (Clerky is the easiest for startups)
• Domain and trademarks: do a basic search for conflicts on your name
• IP agreements signed with all founders and early contributors
• Privacy policy exists and is accurate (what data you collect, where it goes)
• Terms of service exist

Open source specifics (if applicable)#

• License file in the repository (MIT, Apache 2.0, etc.)
• All dependencies have compatible licenses with your chosen license
• Contribution guidelines written

Marketing and Distribution#

Before launch#

• You know exactly which channels you're launching on (HN, Product Hunt, Twitter, both?)
• You've written the launch post and had someone outside your company read it
• You have a personal Twitter presence — people will check your profile
• You have a GitHub repository with a good README and a few meaningful commits
• You've prepared 3-5 specific DMs to send to people you want to personally notify

Content#

• A demo video (Loom works; doesn't need to be produced)
• 2-3 screenshots for the announcement
• The one sentence that explains what your product is (test this on 5 non-technical people)

Product Hunt specifics (if launching)#

• Gallery screenshots are high quality
• Product tagline is under 60 characters and descriptive
• You have hunters who will upvote early (arrange in advance)

Operations#

User support#

• A way for users to report bugs (GitHub Issues, Discord, email — pick one and commit to it)
• Someone is monitoring this channel during launch
• First response SLA defined (even "we try to respond within 24 hours" sets expectations)

Analytics#

• You know how you'll measure whether the launch "worked" (installs? GitHub stars? Signups?)
• Basic analytics are in place to measure this

Communication#

• You have a way to communicate with users post-launch (newsletter signup, Discord, GitHub Watch)
• You know how you'll handle negative feedback publicly (your response tone matters)

The Things We Wish We'd Done#

Set up Discord before launch: Our community discussions started happening on HN and Twitter, scattered. Having a Discord ready would have concentrated the conversation and the community.

Prepared canned responses for common questions: The same 10 questions came up repeatedly in the first 48 hours. Having prepared answers saved time on the repeat questions but we could have been faster.

Had a follow-up strategy: The launch is 36 hours of intense attention. The week after, that attention has mostly dissipated. We hadn't prepared a "week two" cadence — more content, new feature announcement, HN comment thread engagement — that would have extended the attention window.

Frequently Asked Questions#

When is the right time to launch?#

When a real user (not you) has used your product to solve a real problem without you helping them. That's the bar. It's not about perfection — it's about demonstrated value.

Should you do a soft launch before the main launch?#

Almost always. 10-20 users in a private beta period will find the problems that would be embarrassing to find publicly. The bar for the soft launch is lower; the bar for the public launch should be higher.

What if the launch underperforms?#

Re-examine the messaging, not the product (usually). Many products that eventually succeed had quiet initial launches. The Show HN with 15 points that goes nowhere is a signal to improve the story, not necessarily the product.

How long should you wait between launches?#

Product Hunt gives you one major launch, but you can have multiple. Hacker News you can post to multiple times if you have genuinely new things to show. "Show HN: I added X to Y" is legitimate if X is substantial. Don't re-launch the same thing.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →