Spent the week building internal tools and fighting infrastructure. None of it demos well. All of it compounds.
Unglamorous weeks are where the company actually gets built.
Something I unlearned this week. Speed at a startup isn't recklessness. It's accepting that shipping wrong usually costs less than shipping late.
Still uncomfortable. Shipping anyway.
Weekend rabbit hole. How small can a model get and still be useful for real work?
Looking for benchmarks people actually trust. The vendor ones all agree that everything is amazing.
Everyone is giving agents more tools. The bottleneck isn't tools. It's teaching.
An agent with ten tools and real judgment beats an agent with a hundred tools and a prayer.
We decided we have to move faster. Not more hours. Less ceremony.
Founders who came from engineering jobs, what big-company habit was hardest for you to drop?
The more time I spend around sensitive data, the more interesting small local models get.
Some data should never leave the building. A small model running where the data lives beats a big one that needs the data shipped out.
A year ago I'd have called this impossible. Today it ships.
If you're a founder losing sleep from a customer asking for soc2 or to a security questionnaire, this is what I wish had existed when I was you.
SOC2 done entirely through claude code.
Claude Code is already open on every engineers screen.
So I taught it to run your compliance.
Agents do the readiness work with real data from your stack. You run your startup.
Connect the MCP โ watch your audit prep collapse into a workday.
Introducing Blue Magma.
Been chewing on how to eval an agent's decisions instead of its outputs.
Grading the answer is easy. Grading the which-tool-when-and-why is the part nobody's cracked. That gap is where agent reliability actually lives.
This week we're mostly building tools nobody outside the company will ever see. Internal apps that make us faster at helping customers.
Quietly convinced it's the highest-leverage code we'll ship this month.
@i_mika_el We charge from day one. What we obsessed over was making the UX and the pricing both feel easy. Most of the friction people blame on charging is actually onboarding pain in disguise.
@sridharfyi Quite exactly a month ago. Blue Magma wasn't going to become what i wanted it to unless i was 100% in. A month in, that's even more obvious than i thought.