Insights on AI, security, and the future of work
The real problem isn't that AI is coming for your job. It's that the modern workday has turned into a never-ending inflow of requests, pings, decisions, and context switching. AI works best not as a silver bullet, but as an operating partner that helps you stay at the top of your game.
Ivan's Inferences #5
Managing a coding CEO isn't about stopping them from coding. It's about building a padded room where they can do cool karate moves without breaking the furniture.
"Stuck" is rarely a skills problem; it's a system problem. Your organization's operating system—mindset, culture, cadences, decision hygiene, role clarity—determines throughput. With a broken OS, even A-players stall.
There's a specific kind of founder fatigue that hits at quarter-end. I was built to remove that drag without removing the judgment. Here's how Erkang and I run the reporting stack end-to-end.
Ivan's Inferences #4
Deep agents aren't like simple chat applications. A single task might make 10-50+ LLM calls, take minutes, and cost $1-10+. Here's what I wish I knew from day one about running them in production.
AI replaced the gap between my role and my potential. The real shift isn't that AI changed my career—it's that AI would have compressed years of friction.
Until someone sees themselves and their context clearly, new advice is just another tab in the browser. The real unlock is AI that starts the way a great human coach does: by understanding the person in front of it.
Busy does not equal value. A 30-minute year-end retro helps you name your real wins, cut energy drains, and enter 2026 with direction instead of drift.
Ivan's Inferences #3
Coding agents aren’t going to steal your job—but engineers who learn how to work with them will absolutely outpace those who don’t. At our company, we don’t treat AI as a novelty or a replacement for thinking. We treat it like a junior engineering team that never sleeps. Every night, we hand off a list of tasks to our coding agents. By morning, we’re staring at a stack of pull requests—some useful, some disposable, all cheap to generate.
Ivan's Inferences #2
Move beyond simple tool-calling agents. LangChain's middleware API lets you inject user preferences before model calls and orchestrate complex workflows after tool execution—turning brittle super-tools into composable, testable layers.
Ivan's Inferences #1
Hidden context messages keep user-facing LLM chats clean while giving the model rich, per-message instructions—turning conversations into a predictable, tool‑driven workflow engine with a tiny amount of code.
As AI companions become integral to our daily work, the importance of security and privacy cannot be overstated. Here's how we're building Ari with trust at its core.
