Vito Botta

Update: Cancelled Ollama Cloud after it got too unreliable. Switched to Kimi K2.5 coding plan ($39/mo) tonight and it's been brilliant - stable and fast. Need open source LLMs so I can use my preferred tools (OpenCode, BoltAI), but Kimi works for now. Using their per-token API for non-coding tasks.

Upgraded from BoltAI v1 to v2 last week. Global shortcut brings AI over any app, instant workflows, inline editing, and customisable everything from prompts to plugins. I can switch between models instantly and use proprietary or open source models with my provider of choice. Snappy native Mac app, not a browser wrapper. One-time purchase, actual privacy. Finally fits my workflow properly.

The Ollama Cloud subscription is genuinely ridiculous value when it works. I've burned through 949 million tokens on the $100/mo plan using GLM 5, still nowhere near the limits. Mostly fast with minor slowdowns, but today was very slow and sometimes unusable after they added Qwen 3.5. Still impressed, but if reliability doesn't improve, I may switch back to Claude. Cancelled two Claude Max subs ($400/mo) because these open source models are just that good.

Opus 4.6 spent six hours stuck on a Jira issue. GLM 5 fixed it in minutes.

The performance gap? Minimal. The price gap? Huge.

Maybe we should stop equating expensive with better.

There's a special kind of happiness in deleting code. Not just a few lines, I'm talking about hundreds of lines. Whole files. Ancient features that someone built in a hurry years ago and never touched since, held together with sticky tape and prayers.

The best code is the code you don't write. But the second best? The code you have the courage to delete.

I did some digging on "market before you build" after watching founders launch for years. It works when your risk is market risk, not product risk.

When I launched SpringPulse, I'd already built it for a year. I made it for myself at Brella, fed up with bloated tools and stupid simple boards with simulated sticky notes but no insights or analytics.

Know which game you're playing.

Been using Kimi K2.5 heavily and I'm a genuine fan. Beats Claude Opus on HLE (50% vs 32%), costs ~1/8th the price, and actually fixes tests instead of suggesting we delete them. Remembers my instructions without constant reminders too.

OpenCode desktop + Zen inference makes it fly. Curious to try the just-released GLM 5. At this rate, Chinese models will make American pricing hard to justify.

21 years with Rails and I still love starting new projects. Nothing else lets you build this fast.

It's genuinely the only framework where one person can build something properly complex without getting stuck. Rails 7 and 8 have only made this more true.

Honestly gutted it isn't more popular. Watching teams struggle with exhausting JavaScript setups when they could just ship working products quickly.

Here's to the next 21 years 🚀

30 years of coding means I don't "vibe code" (blindly accepting AI output). I vibe engineer (https://tinyurl.com/2kwyk6bh): I design, specify precisely, and review everything carefully. Research confirms professionals "don't vibe, they control". AI amplifies my experience; it doesn't replace my judgment.

Category domains like ai.com are a branding trap. Advertise "AI.com" and you promote every competitor too. Traffic remembers "AI," not you.

ChatGPT hit 100M users without ai.com. They built branded search volume instead of renting generic terms.

The domain bounced between OpenAI, xAI, DeepSeek redirects for years. Owner just leased it. Not a foundation; a billboard. Spend on specific intent instead.

Did some research out of curiosity. Forbes puts Elon Musk at $852 billion. Statistics Finland says median Finnish household net wealth is €96,000. That means one person's wealth equals roughly 8.3 million typical Finnish households - about three times the total number of households in the entire country. He added $400 billion in about 14 months. The numbers have genuinely stopped making human sense.

Not worried about setting up OpenClaw safely. I'm worried about giving any AI model full access to my machine when prompt injection is still unsolved. 900+ instances already open on Shodan. People buying Mac Minis they didn't need.
The trust isn't there yet. I'll wait.

@coderCyclist@infosec.exchange I'm putting the feature live now. I had to sort out a few small issues and fix an IDOR security problem, but overall the amount of work done this fast is really impressive.

Gave Claude Opus 4.6 one task: build a full Jira integration for SprintPulse.
30 minutes. One shot. Auto-accept the whole way through.
I haven't even reviewed the code yet and it just... works.
This is the first feature I've ever let AI build completely unsupervised. We're in a different era now.

Heroku officially went into maintenance mode this week. No new features, no new enterprise sales. Salesforce is moving on to AI instead.
It changed how a whole generation of devs thought about deployment. Sad to see it end like this, but honestly, most poeple moved on years ago.
Choose platforms where you are the customer they are trying to impress - not an afterthought funding someone else's pivot.

Unpopular opinion: most founders are building launch platforms or "ship faster" tools. Those markets must be completely oversaturated. Meanwhile boring industries with real budgets are barely touched by modern software. Best opportunities are where nobody's posting about them.

Hey guys do you mind upvoting my app SprintPulse on MicroLaunch? It would help a lot https://microlaunch.net/p/sprintpulse

I just launched SprintPulse 🎉- a retrospective tool for teams that actually want to improve, not just tick a box.

Real-time collaboration. Anonymous feedback. AI that surfaces key topics, generates summaries, suggests actions, and tracks sentiment + improvement metrics across retros.

Built for small teams, priced for small teams.

Free to start. Code LAUNCH = 30% off forever (first 100 only).

#buildinpublic #buildinginpublic