CodeNote.net

Spoke at TechLead Conference 2026 powered by connpass

#conference #ai #technical-debt #tech-lead #event
Tadashi Shigeoka ·  Wed, April 22, 2026

I spoke at TechLead Conference 2026 powered by connpass on “Tackling Technical Debt in the AI Era.” This post recaps the talk, social media reactions, off-the-record speaker conversations, and a personal connection to connpass that made this invitation especially meaningful.

Event Overview

This was the inaugural tech-lead-focused conference, hosted by Eight (a Sansan product) with special cooperation from connpass (operated by BeProud, Inc.). Being invited to speak at the very first edition was an honor.

The Talk: Tackling Technical Debt in the AI Era

As AI coding agents like Claude Code and Codex take a central role in implementation, the way we frame and repay technical debt is shifting. The talk was organized around four points, drawn from what we actually practice at ROUTE06, Inc.:

  1. Start by asking: “Whose debt is this?” The same code looks different as debt depending on whether an AI agent is navigating it or a human is reading it. For AI, the debt is “context boundaries and the absence of an SSOT,” addressed by establishing a single source of truth, searchability, and right-sized granularity. For humans, the debt is “readability, naming, and structure,” addressed by naming, function decomposition, comments, and cohesion. Humans and AI agents call for different remediations.
  2. AI-first and Human-first repos coexist inside the same company. An AI-first repo (new products) is built assuming AI writes the code, focusing AI on what it does best so debt is harder to create in the first place. A Human-first repo (existing products) resists context pollution and is strengthened along four pillars: faster feedback loops, an agent quality foundation, redefined code review, and a renewed development cycle. The tactics differ, but the target state is the same.
  3. Whether to repay is decided by a template. Two decisions only: “Whose debt is this?” and “Will this hurt us today or tomorrow?” If it does not hurt, wait until the last responsible moment, or wait for AI to make the debt irrelevant. Not every debt needs to be repaid in a hurry.
  4. What actually works is classical discipline. Design intent (what to protect, what to concede), architecture (boundaries, responsibilities, dependencies), and SSOT (truth in one place). If the skeleton is sound, debt can be refitted. If the skeleton is absent, neither humans nor AI can save it. These fundamentals matter more in the AI era, not less.

Reactions on Social Media

Many attendees shared their thoughts on X with the #TechLeadConf2026 hashtag. Seeing how the talk landed with the audience was incredibly valuable. Here is a selection of reactions (in Japanese).

Thank you to everyone who shared their impressions and quoted from the talk.

Reflections

Being invited to the inaugural edition of this event was an honor. The theme and speaker lineup were clearly well-designed from the outset, and being part of such a carefully crafted conference from its very first edition was a meaningful opportunity.

The off-the-record conversations among speakers were equally valuable. Hearing how each company actually adopts AI in practice, the unvarnished version that does not make it into public posts, is exactly the kind of thing that only happens when you show up in person.

There was also a personal connection that made this invitation particularly meaningful. I rely on connpass almost daily, but my history with it goes further back. When I was a junior engineer, I attended Python mini Hack-a-thon, organized by members of BeProud, Inc. (the very company that operates connpass). Being invited years later to a conference where connpass itself is a special cooperation partner felt like a long arc quietly closing.

That’s all from the Gemba of TechLead Conference 2026 powered by connpass.