Blake Xu (@BlakeXu_) header
Blake Xu (@BlakeXu_) profile picture

Founder | AI | Digital Marketing | Growth Marketer 🧾 PoundKit https://teapaw.org/t.co/F0mwtkh0nJ 🏭 FactoryRoster https://teapaw.org/t.co/WQrLacZxgH

Joined September 2026
Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)Photo by Blake Xu (@BlakeXu_)
Before you call your AI product profitable, count: Inference. Retries. Free users. Infrastructure. Support. Revenue screenshots leave a lot out.
1
2
44
Looking for builders working on something specific: A painful workflow. A tiny market. An overlooked customer. A tool you wanted for yourself. What are you making, and who needs it?
1
5
92
A model winning a benchmark is interesting. A model completing your actual workflow reliably is useful. Those aren’t always the same model.
1
7
115
The most useful AI evaluation might be a folder of cases where your product embarrassed you. Keep adding to it. Run it before every model switch.
5
10
245
An AI feature worth paying for: You give it a messy input. It produces something usable. You don’t spend the next 20 minutes fixing it. Simple pitch. Difficult product.
2
1
10
227
Before buying ads, watch someone try to sign up. You might already be paying for traffic that your onboarding throws away.
4
1
14
304
Open source is more than a cheaper alternative. It gives builders the option to inspect, modify, self-host, and keep going when a vendor changes direction. That option matters.
9
12
282
If agents become a major way people use the web, what should websites expose first? Clear documentation? Structured data? An API? A way to pay? Feels like a web design conversation worth having.
6
7
120
A useful web dev exercise: Open your app on an old phone. Use a slow connection. Start with a fresh account. Now try to complete the main task without forgiving anything.
4
10
237
There’s room for more software that never becomes a startup. A tool for your family. A dashboard for a local business. A script for a colleague. Useful is a perfectly good outcome.
7
10
289
Before adding another agent, try giving the first one a smaller task.
4
17
340
Your landing page says “save time.” Show me the task. Show me the old process. Show me the result. Let me decide how much time that’s worth.
6
14
344
Your first 10 customers don’t need to believe in your vision. They need a problem solved badly enough to try something new.
11
1
19
376
A practical way to learn agents: Build one that uses a single tool. Read every tool call. Make it fail. Add recovery. Then give it another tool. You’ll learn a lot before you need a framework.
6
13
280
The AI products I want more of: Less “ask me anything.” More “that tedious thing you do every Friday? It’s handled.”
4
14
309
“Someone could build this in a weekend” is rarely the whole story. Could they find the customers? Earn their trust? Handle the edge cases? Keep them coming back?
7
14
331
Founders: what’s one thing users consistently pay for that other builders keep dismissing as “too simple”?
3
1
10
219
A good coding-agent instruction file should answer: How does this repo work? What should never change? How do we check a change? When should the agent stop and ask? Write it for the next contributor, too.
8
14
386
A coding agent needs more than access to your repo. Give it: A clear task Examples of good code Commands to verify its work Explicit limits The prompt is only part of the setup.
8
7
324
If your product only makes sense after a 20-minute explanation, try rewriting the homepage before rebuilding the product.
8
12
398