CTO at Firmhouse. Please consider the environment before running your business. Join the club.

Joined September 2007
781 Photos and videos
Pinned Tweet
Product/company tools we're using @firmhouse for day-to-day stuff that could fit in one tweet. (Probably) most used first: - @linear - @SlackHQ - @intercom - @github - Google Apps (email, calendar) - @basecamp - @heroku - @NotionHQ - @AppSignal - @CircleCI - @pingdom
4
25
Michiel Sikkes retweeted
Commerce is where this becomes obvious. The moat isn’t the model. The moat is the learning loop built on top of merchant context. Every interaction makes Sidekick smarter. Every outcome creates new signal. AI with context is useful. AI with context that compounds is transformative.
19
8
193
39,182
Michiel Sikkes retweeted
I calculated Tesla FSD's reaction time
340
1,484
7,707
1,326,469
When is valuemaxxing going to be trending?
1
29
People are surprised at LLM models one-shotting full games, or at least 3D engines. Like Minecraft. While indeed impressive that a model can now do that in a matter of minutes. You should realize that this is not *new*. Models are perfectly good at things they already know or can simply pull from the internet. “Build Minecraft” is an easy prompt. Because Minecraft has already been built. Imagine what the prompt and the (creative) work would have been like if you needed to build Minecraft, when Minecraft didn’t exist yet. That’s where your money and awe should be. Building things that already exist is easy. It’s just the compression of time that’s now impressive. But hey, computers can do things fast!
1
2
62
Can we already phone call / voice mode with a remote Codex thread?
27
Michiel Sikkes retweeted
I just bought a Tesla Model Y Full Self Driving is one of the most impressive pieces of technology I’ve ever experienced. I now wish I’d got it a year ago.
420
340
6,408
8,581,146
Michiel Sikkes retweeted
The productivity of working 1 minute every 30 minutes for 16 hours (i.e. 32 minutes) steering agents is much greater than working 8 hours a day (i.e. 480 minutes)
47
12
434
33,712
Michiel Sikkes retweeted
stop obsessing over which model to use. They're all great. The app is what shapes your thinking. Pick the one that makes you reach for it.
9
8
54
2,890
Multi-user agent harnesses are the next best thing.
19
Michiel Sikkes retweeted
Manifesting a summer like Hiroshi Nagai’s artwork
71
2,121
20,702
1,472,809
Michiel Sikkes retweeted
Replying to @t_blom
This problem will naturally tend to go away as companies are grown from the start using AI. Then you don't need to extract any domain knowledge from people's heads; it will never have been in people's heads.
101
67
1,694
221,827
Michiel Sikkes retweeted
When Shopify stock dropped 80% from all-time highs some employees were left with stock options that were essentially worthless. So @tobi completely rebuilt Shopify’s compensation structure in a way no other company does: letting employees choose how much they want to be paid in stock, RSUs, cash, and even Shop Cash. “ You can change it every quarter. You decide how much money you want.” “ You can even use a tool to lock in the value of the stock you receive for three years.” “ You have full agency and you make this choice.” “ It's very popular.”
34
54
1,864
447,650
Michiel Sikkes retweeted
So proud of this new site. Straightforward, readable, calm. You should go check out everyone else's website in our industry. The usual suspects. Can anyone tell the difference between them? It's like they've been all been sucked into the same black hole and hit their own singularity. No differentiation. Identical messaging. Wild claims of helping you achieve a 24/7 AI work fantasy. Really? There's that much work that needs to happen behind your back while you're sleeping? Since when? And who created that illusion? Agents with creepy masks and emotionless faces? What? The cosplay is something else.
May 26
Basecamp 5 is live! And the new website is a throwback to a simpler time. Just show the damn product. Just list the features. Yes, it's agent accessible, but it's not agent hysteric. basecamp.com/
46
19
303
51,513
Similar. Context loading is key for it to build the right thing. This is not about plan-mode. It’s about making sure all relevant bits and pieces are in context before you actually ask it to build. Or build whole your in this mode. Building > Planning
my typical input to codex is like: "okay I want to achieve X. but I can think of like 4 ways of doing it. if we do Y we have to deal with Z. if we do A we have to deal with B. I think? maybe? idk what I'm talking about. is there some way to do C without D? what do you think?"
44
I said the following in slightly different 3 months ago. Which is a core reason why (from a business value perspective) I don’t see the need to “write code” anymore in the coming months.
Replying to @icanvardar
I think the treshold for clean code has shifted now agents can maintain messier code more easy than humans. Speed at business value was always more important than clean code. But value speed decreases if you don’t (or didn’t) keep it clean. It’s not that clean code isn’t important anymore. But the criteria should be maintainability. Not clean code in itself. Agents can maintain messier code more easily now. So clean code has become less important. Clean code can still be wonderful in itself. But more as a form of art. Not as a form of necessity. Teach your agents to write clean code. And you’ll have the best of both worlds.
1
1
95
RT @theo: A lot of people are building with the assumption that the codebases we work in today will still matter next year. I’m not sure i…
64
I’d say we’re almost right on schedule. Maybe 1-2 months delay. I like to move fast
This year (probably closer to next 3 months) I think programmers should not see themselves as programmers anymore. Not even the kind that uses AI agents to “generate code” - that’s still programming, that’s still thinking your job is to “build stuff with characters on a screen”. I think this year programmers will more and more become “systems guardians / maintainers”. Guiding and guarding that what the end user / customer needs happens, in a dynamic environment where agents will generate most of the code, with real humans hardly inspecting it anymore. This means that your job as programmer will be creating / instructing agents and building workflows and environments where these agents will do it safely. And steer them left or right and nudge them where they go off the path. Keeping systems maintainable and reliable is your new job. In an environment where problems are solved at the speed of light. We used to care a lot about “well factored code” because humans had to maintain them. The human factor will mostly become obsolete. While I do think that a lot of maintainable code best practices are critical for agents to also keep doing their job. The importance of the “craft” of writing code for the humans may become obsolete very fast. The best programmers will know how to see the right path. And they will be the people who can instruct and teach agents at scale. But agents and LLMs will be the primary “audience” of writing or reading code now. So we should change our perspective of “what writing good code” means. Maintaining the stability, effectiveness, speed, etc of the whole system is now the job of programmers. Without perhaps ever writing a single line of code again.
510
Your job is context switching. Not coding.
62
We’re taking big agentic and workflow leaps at @firmhouse next week. But we are also deploying a stream of Quality of Life updates continuously. We’re now rolling out an update to our reports module. Generating long reports should now be faster, more stable, and we’ve added an indicator on its export progress.
60
Michiel Sikkes retweeted
It's wild how quickly after going full FSD, every other car feels broken. It's like coding with AI, you can never go back.
42
8
244
26,646