We all tell LLMs what to do - research, implement, analyze.
Something I started doing after first instruction/implementation is done is to have it assume it's all wrong.
To use ever skill/agent/tool at its disposal with the assumption the initial implementation is wrong, outdated, doesn't make sense, inaccurate, lacks research and best practices.
The point is to have it prove the initial implementation achieved the instructed goal. Works very well so far.
Quick tip: LLM-based E2E testing
When you want your coding agent to perform full end-to-end testing (cypress kind of thing), send it to inspect the entire codebase and create a directory under docs/e2e to create markdown files describing each scenario in the user-flow, elements involved and expected behaviour.
Every interactive element should have a purpose, objective and expectation/success criteria. Create a README.md under /docs/e2e/README.md that indexes all the markdown files in a table.
Better yet if you instruct the agent to create file per app's section to not overwhelm the LLM.
Today I built a demo for a prospect client from the live transcript as the meeting was in progress. I shared the link to the demo at the end.
Ooenclaw Demo builder agent connected to my meeting app. I asked the prospect exactly what the builder will use making sure it has enough context for the demo, it suggests me questions through Telegram to make sure I get all necessary information for next steps.
After the meeting is created a full product spec and further questions to clarify MVP are documented.
Anyone can guess if I got the client?
How is this different from any other marketing strategy? Isn’t it because it’s software related?
I understand what you’re saying although I don’t understand the need to throw shade, but suggesting to ignore people because they present a different approach or to your eyes a scammy way of marketing is a bit dangerous.
For people to understand what is and isn’t real, they need to see both sides. That is actually what divides people, shutting down everything that we disagree on.
People are smart or life will teach them what is real, at the end they will decide for themselves.
I agree if you’re talking about going from zero to “complete salesforce”, it’s not realistic. But you can go from zero to MVP, start getting users and evolve the app.
For people who don’t know how to build software they just need to get it off the ground then as income comes in hire someone who can scale it.
@businessbarista Hi Alex, I do this as well. I wonder your thoughts on an app for this. I created it for myself but thought it would be useful for others as well.
https://t.co/sLLMmARoXd
Seems the only issue is direction. Maybe you haven’t though in terms of long-term vision. Everything you’ve learned is enough to build that vision.
IMO is not about being the absolute expert in a field, that’s just accumulated knowledge, but what will you do with that?
Building something that aligns with your life goals, it’s the only way to achieve satisfaction with your progress.
What would that be? Is something you must figure out on your own by including you entire life goals in the equation.
Learn as you build.
My two cents. Wish you the best.
Doesn’t this post implies ASI it’s something that is reached and it stays that way?
I agree with your assumptions although why would ASI be static as opposed to having recursive learning ability and even more importantly to take actions with self-given goals.
We get used to things that are static or change slowly enough to give us time for adaptation. Plus ASÍ wouldn’t just happen, corporations would stop advancing once they get to a good enough point where they know they can still control it and have achieved their goal.
Hi Zara. Looks awesome and very useful.
One question, if you were to market it, how would you handle content security when process through the LLM providers?
I have a similar use case without the video and for an entire different use case but I stumbled across this question often. I thought about open source models but then it becomes a matter of hosting and mild technical settings that some users might have issues with.
Is this even a concern or basic auth with terms of service about content being processed through 3rd-party providers would basically be it?
Thanks for sharing.
I've intensively used Gemini-3, it's awesome for implementation, a bit weak for crafting documentation like PRD, technical documents, and brainstorming.
Verdict so far:
- Gemini 3 - For building
- Claude Sonnet-4.5 & Opus-4.1 - for brainstorming and crafting docs.
Maybe create MCPs around it to be connected to any LLM client and would probably end the wrappers market.
On the other hand it’s not the tools that provide actual value, they’re just the medium, the trick is to combine them to solve specific use cases, and for that you need to understand the problems they’re going to solve.
It comes to a point (if we’re not already there) where industry experts are more valuable than technical ones.
@Pauline_Cx Wish you told me 7 years ago when I was launching my first project thinking “I must be prepared, people will be flooding my awesome app…any second now”.
💀⏳