San Francisco, CA, Dec. 09, 2025 (GLOBE NEWSWIRE) — Deus Ex Machina Capital announced the release of two research papers examining why AI-assisted productivity gains are diverging across organizations. The papers were authored by Francesco Bisardi, an independent researcher who leads the firm’s research program. The work argues that AI agents and modern developer tooling can create step-change execution speed when teams redesign workflows rather than layering AI on legacy processes.
Bisardi said:
“Over the next five years, the binding scarce asset will not be access to UI-based chat models, but the capability to redesign human workflows into agentic systems and apply them to mispriced surfaces still anchored to human labor or legacy R&D cost assumptions.”
The first paper, “An Approach to High-Velocity Development through Systematic Context Engineering: A Case Study,” introduces a context architecture designed for fast AI-native software development. The approach, based on a four-pillar framework, combines:
- A declarative rule layer that encodes architectural, process, and quality invariants
- A programmatic Repo Model Context Protocol (MCP) layer that exposes live project structure, data, and tools to AI assistants and agents
In roughly fifteen part-time weeks, two researchers built a production-grade research artifact (Bizie.app) that reproduces a representative set of features found in mature event-productivity software, illustrating the practical implications of the proposed context architecture. The paper serves as a case study and a practical guide for aspiring entrepreneurs, clearly defining the distinction between vibe-coded prototypes and production-grade applications.
The second paper, “The LLM Productivity Cliff: Threshold Productivity and AI-Native Inequality,” addresses a central tension in the AI narrative: widespread access to LLMs has not produced uniform productivity gains. Based on early empirical evidence, the researchers argue that:
- There is a threshold productivity level below which additional AI assistance has limited impact and above which returns become sharply non-linear.
- The decisive variable is AI architectural literacy: the steepest gains accrue to teams with the skills to re-engineer human and UI-based workflows into agentic systems, while organizations that simply add AI tools on top of legacy stacks and processes see only marginal improvements.
Press inquiries
Deus Ex Machina
https://www.deusexmachina.capital/
Press
press@deusexmachina.capital
San Francisco
