# Writing · Lokesh Saini

> Canonical (HTML): https://www.lokeshsaini.com/blog
> Markdown twin of the page above. Contact: hi@lokeshsaini.com · Last updated: 2026-06-18
> Profile: https://www.lokeshsaini.com/llms.txt · API: https://www.lokeshsaini.com/api/portfolio · Resume: https://www.lokeshsaini.com/resume/Lokesh_Saini_Resume.pdf

- [Complexity Is a Safety Problem](https://www.lokeshsaini.com/blog/complexity-is-a-safety-problem) (2026-06-15): Feature count debates miss the point. When your users are vulnerable, every added setting is a new failure mode, and cutting scope becomes a form of harm reduction.
- [The Workshop That Changes the Project](https://www.lokeshsaini.com/blog/workshop-that-changes-the-project) (2026-04-16): Most workshops produce sticky notes and good vibes. A few produce one insight that reframes everything. The difference is the setup, not the activities.
- [Prototyping in the Wrong Tool on Purpose](https://www.lokeshsaini.com/blog/prototyping-in-the-wrong-tool) (2026-04-16): The right prototyping tool depends on the question you're trying to answer, not the phase you're in. Sometimes paper beats Figma and code beats both.
- [Why Node Graphs Fall Apart at 15 Nodes](https://www.lokeshsaini.com/blog/node-graphs-fall-apart-at-fifteen) (2026-04-16): Node graphs demo beautifully with 5 nodes. At 15 they become unreadable. The problem is topology, not tooling.
- [Designing a Learning System Around How My Brain Actually Works](https://www.lokeshsaini.com/blog/learning-system-for-my-brain) (2026-04-16): I have ADHD. Most learning tools ignore that. So I built one that doesn't.
- [Designing for Split Attention: Two Screens, One Story](https://www.lokeshsaini.com/blog/designing-for-split-attention) (2026-04-16): When your interface lives on two devices, the user can only look at one. Sound, haptics, and invisible latency absorption become your real design tools.
- [I Built an Anti-Sycophancy System for My AI Coding Agent](https://www.lokeshsaini.com/blog/anti-sycophancy-system) (2026-04-16): When your AI assistant agrees with everything you say, including the things you get wrong, you have a compliance problem. Here is how I fixed it with hooks, rules, and adversarial agents.