How to build your own AI agent
Trials at FidetoLabs, and case studies from other companies.
Latest posts
4 min read
Your Agent Is Reading Files You Never Told It About on Substack
Why giving your trading LLM file system tools is a terrible idea, and how to fix it with an API.
7 min read
How Claude Code’s Subagents Keep the Main Context Clean on Substack
Why the useful thing a subagent does is not “run in parallel” but “do the noisy work somewhere else and hand back only the summary.”
6 min read
How Manus Keeps Tools Stable by Masking Logits Instead of Swapping Them on Substack
Why a growing tool catalog breaks long agent loops — and how Manus constrains the next action without rewriting the tool list.
6 min read
What Higgsfield Supercomputer’s Orchestrator Actually Does on Substack
How one brief gets split into steps, routed to different models, and run without you choosing Claude, Seedance, or Kling each time.
4 min read
The 130% Growth Formula: Inside the Revenue Machine of Korea’s First AI Unicorn on Substack
Evaluating the financial tension between soaring corporate adoption and the structural deficit of a top-tier large language model developer.
5 min read
How Korean Deep Tech Solves the Compute Bottleneck for Physical AI on Substack
Looking past consumer software to explore the optimization layers required for real-world robotics deployments and localized autonomous systems
4 min read
How a Korean Startup Rules Logistics on Substack
South Korea’s hidden Physical AI frontier.
4 min read
How to Bypass Traditional B2B Distribution by Embedding Directly into IDEs on Substack
Inside the product-led strategy that bridges the gap between deep developer workflow and enterprise software adoption.