To Copilot and Beyond: 22 AI Systems Developers Want Built
Central argument
Through empirical research with software developers, this paper identifies 22 distinct AI system archetypes that practitioners actually want built — going well beyond the code-completion paradigm exemplified by GitHub Copilot. The central finding is that developers envision AI not merely as a productivity accelerator for writing code, but as a collaborative system capable of handling complex cognitive tasks across the full software development lifecycle: architectural reasoning, debugging, documentation, onboarding, and cross-team coordination. The paper argues that current AI tooling addresses a narrow slice of developer need, and maps a more expansive design space for what purposeful AI assistance in engineering could look like.
Critique
A core tension in this work is that it captures what developers say they want, which is not always a reliable proxy for what would actually improve outcomes at the system or product level. Developer preference research is prone to reflecting individual productivity fantasies rather than organizational or quality constraints — a developer might want an AI that autonomously refactors legacy code, without accounting for the governance, accountability, and review structures that exist for good reasons. The paper risks conflating desirability with feasibility or wisdom, and a stronger version would cross-validate the 22 archetypes against evidence of where AI assistance demonstrably improves team-level or product-level outcomes.
Why it matters for product
For a CPO, this paper reframes AI investment decisions: rather than defaulting to developer tooling as a productivity play measured in lines of code or cycle time, it surfaces a richer taxonomy of where AI can intervene in the software creation process — from discovery and scoping to cross-functional coordination — which maps directly onto product team design and where to allocate AI capability budget. It also has strategic implications for product leaders building developer-facing products, since understanding the 22 archetypes provides a structured framework for identifying underserved segments and defensible product territory well beyond what Copilot-style incumbents currently occupy.