Builder Journey - 1

I’m back at square one, and I’m honestly pumped and excited about it. I remember being superbly passionate about an idea and working hard to make it real—and I can feel that same energy now. Why am I doing it?

The answer is to challenge myself. A lot of people, including me, believed that SaaS applications were dead. The reality is very different. SaaS revenues are higher than ever and can be cracked if the right problem, the right product solution, and the right creative messaging are executed well. If you can crack any two, you’ve got a strong, investable business; cracking all three gives you a godlike superpower of going direct to consumers. With AlmondCoder, I feel it hits all three (for now).

What is AlmondCoder?

AlmondCoder is a programming tool similar to Cursor—basically, it helps you “vibecode” your projects easily. It has features I think are crucial as a builder, such as:

• Run Claude Code / Codex / Cursor CLI in parallel, doing different things. No more waiting for one prompt to finish before starting the next.

• Create a Git subtree for each prompt and merge easily using an interactive GUI.

• Get a simple, editable plan for your prompt that shows what changes Claude Code will make and where the prompt will execute. Often, Claude Code struggles to interpret prompts, producing the wrong plan. This helps you correct it before changes are applied.

• Onboarding prompts for open-source projects that let others discover your project architecture quickly; the generic ones help people learn the architecture faster.

• Prompt Pills: for each project, define recurring context “pills” that are appended to your prompts. Pick the pills you need and go.

• Many features (Prompt Pills, onboarding prompts, auto-merging) are adaptive—parts of the software are generated on the fly based on user prompts. This makes the tool effectively “create software on the fly.”

We’ll eventually allow multiple people to collaborate on the prompt and plan and get it approved before Claude does its magic.

Now, the hard part: executing a tool like this is still a bit complex. That includes designing the landing page, getting the messaging right, building the product, and pushing content across social to drive adoption. It’s not easy—especially as a bootstrapped founder (though arguably easier because the hunger to reach revenue is higher). But even in the age of AI, it’s still hard.

I’ll be documenting my journey in a structured way on this blog for future reference. My goal today is to ensure that:

a prompt creates a new Git subtree
conversation history is saved locally on disk using a specific JSON schema
clicking items in the prompt history loads that prompt’s context
I can create a new prompt with the right scaffolding

I’ll share progress in Builder Journey – 2 and how it’s going. Obviously, there are many finer details associated with the above that still need to be addressed.

 

What jobs will AI do?

If we are able to define what kind of tasks AI will be good at then retrospectively we will be able to come up with what tasks AI is not good at. So what is AI good at? AI is most certainly good at coming up with solutions where the outcome of the task is clearly defined, which is why AI is not going to replace researchers. Research at it's very core does not have the outcome clearly stated, thus the process is more important than the outcome, making it very hard to automate. 

A lot of people say coders, marketers, editors will be replaced but I think these will kind of become a single layer of identity, these people's jobs will converge into obtaining a specific outcome by breaking them into smaller chunks of outcomes. it will still take immense creativity (perhaps more, as lot more is being expected from the same person), thus these chunks of outcomes will be largely be done with the help of AI. 

So much hype around AI. So AI enables the creative, the deep thinkers that are excited about the output more than process. AI will become the process, does not mean you do not have to understand it. You still need it but a larger part will be done by large language models. 

The Only Actress I have a crush on