Voice typing works great for longer texts with at least 1-2 sentences. Sometimes I need to type just one short sentence, STT usually messes up and it’s quicker to just type it.

One would think that Google AI handles such a simple task seamlessly. It's still early before AI is ready for everyday tasks.

Gemini sucks at editing Google sheets.

I've uploaded two sheets from Drive and prompted Gemini to structure and copy data from one file to the other, and it completely failed. Totally useless.

These days, we can say, “I remember when the codebase used to be a real competitive advantage that was hard to copy.”

#buildinpublic

I like observing what drives people beyond just money. For example, the feeling of importance is a strong non-financial motivator that keeps people engaged in their work. Jobs that give people that sense of importance usually have lower turnover.

It makes more sense to worry when there is no competition around.

Demand for critical software isn’t going anywhere. Companies with strong, durable moats will still be around and will likely benefit from integrating AI. Switching costs and other competitive advantages don’t just disappear.

Yes, AI will boost productivity, which could lower revenue per unit of output. And yeah, some business models might shift toward usage-based pricing. But honestly, I think the fear is exaggerated.

It’s crazy how hard software stocks got hit in 2026.
Feels like the market is pricing in a permanent decline in profitability. But I’m not buying that narrative.

I haven’t built it yet, was just talking to a potential customer last week who uses an agent to extract accounting data, and that gave me the idea.

If a potential customer says they’re using an AI agent instead of your product to solve something, think about building an MCP server. Agents rely on tools, so this could be a new channel to reach your customers.

#buildinpublic

Check emails in the morning, social media once a day, and messengers once a day or set up notifications.

I’ve noticed that constantly checking social media, emails, or messages really fuels anxiety. It’s surprising how much better I feel when I break that cycle and limit it.

Don’t overdo online reaching out. It can turn toxic.

#buildinpublic

Just showing up here when the business isn’t working out shows great dedication and perseverance.

It’s easy to fall into the trap of using AI for tasks where it saves little to no time or just isn’t worth it.

Lately I’ve switched virtual pipelines for real ones, and data flow for heat flow.

Working with something tangible in the real world feels really refreshing for those of us in digital.

For many customers, design isn’t that important. They care more about other aspects of the product. And with AI tools improving, the cost of creating high-quality design will keep going down.

I’m afraid that design will matter less and less for product differentiation. Even now some people probably overestimate how much of a competitive advantage it really is.

LLMs aren’t deterministic and struggle with more complex artithmetic calculations. That means their outputs aren’t consistent and reliable.

And users don’t like products that behave inconsistently.

I’ve made that mistake myself.

You don’t want to integrate language models for tasks whose results can be obtained in a purely programmatic way, especially when using arithmetic calculations. Why?

#buildinpublic

I keep running into the same thing: no replies to my messages. Even though there’s usually no quick fix, it keeps proving a simple truth:

The more interest someone has in talking to you, the faster they’ll reply.

Next step: how is it possible to make myself/my product interesting to the recipient?

Solopreneurs can finally use “we” in their product messaging without feeling awkward or embarrassed. They can just casually say their team is made up of AI agents 😁

In other words, sales is about finding people who already have a need, not trying to create that need.

I’m always surprised by how many boring profitable businesses don’t have any social media presence

I’ve found that supervising the AI and making sure we’re aligned on the task design leads to better results for more complex tasks.

The biggest challenge here is the lack of trust, which leads to micromanaging any complex tasks.

If “vibe coding” means coding without supervision, what’s the technical term for using AI coding tools in a coordinated, supervised, human-in-the-loop workflow? What would you call that? #buildinpublic

Yes in this case, higher productivity means skipping some unnecessary tasks

Some decline in coding skills is inevitable. People tend to get lazy and stop thinking as hard when there’s an easier way to do things.

Yes, AI tools sped up the coding and prompting. But they had no real understanding of what the right workflow should actually look like.

There seems to be a tipping point in project complexity where pure AI-driven design starts to fall apart and stops adding any real value. (5/5)