Guido Boyke

Solo founder building Lalabase, an AI-native project tool for tiny software teams. Daily notes from the workbench. Your data stays yours. Improv actor at night.

[Guido Boyke @gboyke]

@wertform_mcm Versuchs mal mit lalabase. Kannste auch selber hosten. Besser als alles nochmal neu erfinden.

[Guido Boyke @gboyke]

What's a feature in your PM tool you turned off within the first month?

Ours: auto status notifications. Every card move pinged the whole team. A week later everyone had muted the channel.

Bonus points if it's the feature that sold you on the tool.

[Guido Boyke @gboyke]

https://t.co/Inj5oaaR1y

[Guido Boyke @gboyke]

@Bjoern_Hev Germany. Klecken.

[Guido Boyke @gboyke]

What belongs to a project belongs in one place.

AI coding will get us there, for a reason that has nothing to do with quality: features got so cheap that every tool swallows the next one.

Bloat included. And the fix: what wasn't expensive to build is easier to throw away.

[Guido Boyke @gboyke]

I'm officially done going to the dentist.
Doing it myself.

On Monday I bought a drill with super tiny bits. And a Dremel to go with it.
Temu had one of those tiny cameras on a tube.

Got the know-how over the weekend via ChatGPT. It's way easier than I thought. No idea why people study for five years for this, half of that is definitely insurance paperwork.

Haven't looked into the anesthesia part in detail yet, but I'm going with the "learn by doing" approach. If it hurts, I must be in the right spot.

Already came up with a name for it: Vibe-Dentisting.

So cool.

Starting with the back teeth. If anything goes wrong, nobody sees it anyway.

A dentist will probably say this is dangerous. Of course they would.

The Dremel cost 34 euros.

Man, the amount of money I'm going to save. I am so euphoric.

I'll keep you posted on how it goes.

.... brrrrrr....kkkrrkrkrkkr....brrrrdkkkr.

[Guido Boyke @gboyke]

@Surendar__05 books, magazines and later online documentation.
And a lot of time to try things out.

and in the old days ökosystem was not that tattered. You only hat Assembler, Basic, Pascal, Fortran, thats it.

[Guido Boyke @gboyke]

@0xKusha Claude, make me rich!

[Guido Boyke @gboyke]

Claude Fable 5.1 is out, and daily life with Claude Code is shifting in ways the press release glosses over.
Token prices look identical, and Anthropic swears agentic sessions will get cheaper. Here is what actually matters.

Longer coding sessions. At high effort—the Claude Code default—5.1 pulls away. Max effort pushes Cursor Bench to 73,4 percent.

The cost reality. Independent audits show tasks can actually cost more because the model burns about 1.7x more output tokens to hit the same result. Token pricing is a vanity metric; cost per completed task is the only ledger that counts.

75 percent off cache reads. Inputs and outputs stay flat at 10 and 50 dollars per million, but cache reads drop from 1.00 to 0.25 dollars. Since Claude Code re-reads the entire context on every step, this is the one real lever for savings.

Quieter behavior. Fewer intermediate thoughts, fewer bundled tool calls, and more roundtrips. For small edits, it prefers rewriting whole files—though the edit tool usually catches it.

Enterprise gets new cloud guardrails, distillation blocks, and EU watermarks. Yawn.

Here is the real talk from my desk:
I code in Fable. I prefer it to Opus by a mile. But sometimes it loops on minutiae and drags things out.
And it is expensive. I am on the highest tier available, and I still hit limits fast. OpenAI’s GPT 5.6 Sol handles a huge chunk of my work for a fraction of the cost with zero noticeable drop in quality. Next year it will be a different name, and the sentence will still be true.
Where Fable earns its keep: code reviews. They come out structured properly on the first pass. No hand-holding required.
Because this math shifts every few months, the AI provider sits in a drop-down menu inside Lalabase, not hardcoded into the source.
We will see if the cache-read discount actually bends the curve this week. I’ll report back.
What is your model stack looking like right now?

[Guido Boyke @gboyke]

Nobody ever got fired for buying IBM.

IBM just goes by different names today: SAP, Jira, Teams. Nobody has ever actually liked them.

If you pick what everyone else has, you're never to blame. The math works out for the individual, but not for the team. One person bears zero risk; the others pay the price every day. Or they only open the tool when someone is watching.

Two clicks here, a mandatory field filled with "tbd", a search bar that can't find your own comment from last week. None of that ever appears on an invoice. And because it never shows up on a bill, nothing ever changes.

It’s the quietest form of waste I know. It looks like a sensible decision—and it is. Just for the wrong person.
This isn’t just a corporate problem. I’ve worked in small companies where this exact reflex ruled. Yet at five people, the alibi doesn't even work. There’s no committee to hide behind. Not even a purchasing department to co-sign. There is just you.

AND EVERYBODY KNOWS IT!

Yet they still reach for it. People with "Better safe than sorry" signs hanging by their desks.

This sounds like a disadvantage, but it’s actually the superpower of being small. You don’t need to cover your ass because the CYA mindset won't protect you anyway. You can choose what your team actually needs instead of what’s easiest to justify.

I’ve never chosen what everyone else has. The trade-off? I have to explain it every single time. Those who pick what everyone takes never have to say a word about it.

Sometimes the tool everyone has is the right one. If thirty people use it and half already know it from their last job, the saved onboarding time beats out any custom fit. Then it’s a real decision. Not just cover your ass.

I sit on the other side of the table now, building a tool nobody knows yet. And "nobody knows it" is a genuine hurdle during adoption. I can't sweep that under the rug by pretending we're already a giant incumbent (I could try, but it's silly).

What I can do is lower the cost of a mistake. With Lalabase, you can take all your data and walk away at any time. Getting in is my job. If it doesn't fit, you aren't chained to it. I can beat the IBMs of this world at that game.

I consider the CYA reflex the most expensive habit in our industry. Change my mind.
When was the last time you introduced a tool that nobody else in your circle was using?

[Guido Boyke @gboyke]

Yeah, you really have to develop a feel for when to use which model. For example, when building a feature, knowing if it's crucial to use something like Fable 5, or if it's worth going with a slightly smaller model.

Another thing that takes practice is knowing how many review loops you should do and figuring out the exact point where extra iterations stop improving things and just start convoluting the code. Those are the kinds of mistakes you make that mostly end up costing you a lot of time.

How to handle this efficiently probably also depends a bit on the programming language and framework. It's the kind of experience you only gain by actually doing the work.

[Guido Boyke @gboyke]

Because people want to build a project they can make money with. That’s point one, and it’s hard to make money with a compiler.

Another reason is that people don't even know what a compiler is; at least, a lot of people who write web code have never even heard the word.

And the third reason is that you still need a certain fundamental understanding to program a compiler, even when coding with an AI, and most people simply don't have that know-how.

[Guido Boyke @gboyke]

Have you ever considered using the maximum subscription at Anthropic and simultaneously the maximum subscription at OpenAI to avoid API costs at all costs? I use the subscription exclusively for coding.

[Guido Boyke @gboyke]

When considering a SaaS solution, is it a selling point for you personally if the vendor also offers an open-source version?

[Guido Boyke @gboyke]

@mvpdhiraj That means one option could be to offer managed self-hosting, for example. So more like running your own installation, but with support. Similar to how it works with WordPress.

[Guido Boyke @gboyke]

If you're building a SaaS, are you also considering offering an open-source version?

[Guido Boyke @gboyke]

Claude Design is really amazing. Especially these interactive views. Just built in no time. It's crazy how much time it saves. But it also really tempts you to "play around" a lot. Have you guys used it yet?

[Guido Boyke @gboyke]

@danellisona What is "sleep"?

[Guido Boyke @gboyke]

Two clicks too many. Reason enough to throw away weeks of work?
The AI chat in my project management tool was working fine. It’s gone anyway.
It sat directly on the ticket. It knew everything about it, and absolutely nothing else.
Except the questions I actually have are never so small that they only concern one single ticket.
"How many hours are still open on this project?" "What did we decide back then, and why?" "Is there anything left over this month that hasn't been billed yet?"
For every single one of these questions, I had to leave the ticket and go somewhere else. Two clicks.
And so, I simply didn't ask them. It felt completely cumbersome.
I run my own client projects with this thing, which means I am my own user. If something feels cumbersome to the person who built it, I don't even need to think about anyone else.
The comfortable route would have been to just leave it in. After all, it works. It looks good on the feature list, and nobody complains about a chat they never open.
This is exactly how applications get bloated. Not because someone needs more, but because someone built something once. In every program you work with daily, there is a menu item that nobody has touched in two years. It's still there because taking it out would mean admitting that it missed the mark right from the start. So it stays, and you end up paying for it.
In previous projects, I've very rarely seen features deliberately ripped back out. The weeks that went into it are gone. They are gone if I remove the chat, and they are just as gone if I leave it in. (And to be honest, taking it out costs time and effort too; it doesn't happen by itself.)
There’s a widely cited term for this: Sunk Costs.
So, back to square one to rethink the concept. I can afford to do this because there is nobody sitting in the room who built the feature and now has to justify themselves. I am that somebody. With Lalabase, a feature isn't judged by what it cost to build, but by whether I actually use it myself. And it turns out, I use things differently than I previously thought.
What feature in your project management tool hasn't been opened by anyone in two years? And why do you think it's still in there?

[Guido Boyke @gboyke]

“Build an MVP.” “Start small.” Usually solid advice.

For what I’m building in Lalabase right now, it’s completely useless.A half-baked AI feature doesn't deliver half the value. It delivers zero.

We currently have a simple chat for testing. It’s neat, but I can find tickets just fine using standard search. A real assistant needs to kill the endless clicking through list views. It needs to answer hard questions:

"Are there any open items left this month?"

"Why did we decide on XY back then?"

You can’t fix that with a clever system prompt. This is deep software engineering. The data layer has to be completely rethought to be queryable, and the permission system must be watertight. The AI cannot leak data you aren't allowed to see in the regular UI. Period.The classic PM objections: Which customer asked for this? Which KPI does it improve? I don’t know. And honestly, I don't care.Iterative value is a total illusion when it comes to AI.

An assistant that answers "I don't know" to 7 out of 10 questions isn't 30% useful. It’s 100% frustrating. The user tries it twice, closes the tab, and never comes back. Below a certain critical mass, the feature is just worthless bloat. You only know where that threshold is once you’ve crossed it.

So, I’m building it. Without permission. Without made-up metrics to offload the responsibility.I run my own business on Lalabase. If I don't catch myself using this chat constantly very soon, I’ll rip it out without a second thought. Who hasn't sat through this exact enterprise theater about "the new feature" in their company? ;-)

[Guido Boyke @gboyke]

@mvpdhiraj My product is a project management SaaS. I'm using the open-source version to remove the fear of vendor lock-in.