Lightning Talk: AI Adoption Without Vast Tech Debt
Software development looks very different today, and I no longer think about tech debt the same way. I group my approach to minimizing AI-era tech debt into three areas: tech, people, and how we ship AI-native software.
Tech #
Coding #
Ground zero for AI tech discussions. Leading coding agents include Claude Code, OpenAI Codex, Cursor, GitHub Copilot, and Gemini Code Assist. Specifics here focus on Claude, where I am currently maxing out three subscriptions.
Unlike humans, agents don’t struggle with deadlines or extra work. The real driver for tech debt in an agentic world is misalignment. Keeping all developers agents rowing in the right direction is key, and the way to do that is via context window management:
- Context windows are 200k to 500k tokens in Claude Code, which are about 500 - 1250 pages of text
- We need to fit everything about our company, policies, vision, market, product, organization, team, project, personal preference, feature specification, and current conversation in this window
- Everything single thing you hope your senior engineers know
- We can host a bunch of markdown from each relevant team and level of the organization hierarchy to ensure all agents have the same context over time
- Agent Skills are one way to accomplish this, implemented as directories with
SKILLS.mdfiles - IT team maintains company policies, SOC-related instructions, handbook. Division leaders maintain… etc. etc.
- Agent Skills are one way to accomplish this, implemented as directories with
Coupled with context window management, the monorepo remains the simplest way to keep context unified across related systems.
Ancillary Tech #
A 10x increase in coding speed requires a 10x increase in everything around it:
- Code review: GitHub Copilot, CodeRabbit
- Testing: coding agents can do this really well, use Playwright MCP, Chrome devtools MCP
- CI/CD, infra, monitoring: anything defined as code the agents do well
- Security: Semgrep, Snyk, Dependabot with cooldowns
People #
People are far more productive, but our human context windows haven’t changed in 200,000 years. To stay on the same page, teams need to be smaller.
- Smaller teams with tighter coordination
- Larger ownership per engineer
With empowered individuals and small teams, it becomes more important to have clear boundaries around systems to maximize autonomy & speed with decentralization. If Amazon made the switch to a Service Oriented Architecture at 7,000 employees, that number is much, much lower today.
Additional tactics to keep people coordinated with AI:
- Agentic access to your ticketing and other SDLC systems via MCP
- Dedicated early adopter(s)
- Clear product processes around specs, reviews, and launches
Shipping AI-Native Software #
The proliferation of agents has come with the need for new types of software systems more usable by agents. Besides Skills mentioned above, we builders are being tasked with building novel MCP and App systems.
MCP #
Model Context Protocol (MCP) lets agents interact with our systems. If skills are the static websites of the agent internet, MCP servers are the dynamic websites.
MCP is about a year old and should be treated as early infrastructure. It will change significantly, especially following its move from Anthropic to the Linux Foundation.
There’s no way to avoid a significant amount of technical debt as the MCP protocol changes out from under us, but we can at least offload some of the work to other people by adopting open-source software. In MCP, a great open-source protocol wrapper is mcp-use.
ChatGPT Apps #
ChatGPT Apps let anyone embed interactive, visual widgets in ChatGPT conversations for use by ChatGPT’s 800M users and ChatGPT itself. If a user is asking about cold symptoms, OpenAI can render the CVS map of nearby stores for medicine.
ChatGPT Apps launched in October 2025. They are MCP-based, but include extra OpenAI-specific UI protocols on top.
Like (and alongside) MCP, this protocol will change a lot and we can offload work to open-source software. For Apps, the leading framework is sunpeak, which I’ve been building the last few months!
Wrapping Up #
The AI era demands a new playbook for managing technical debt. By keeping context windows aligned, teams small, and embracing emerging protocols like MCP and ChatGPT Apps, we can unlock 10x productivity gains without drowning in legacy complexity.
Abe Wheeler is the founder of sunpeak, provider of the leading ChatGPT App Framework. He recently sold his company Trigo, which applied AI to property tech.