Dan Cleary

Dan Cleary is the co-founder and chief executive of PromptHub and Converge, companies building tools to make AI prompts more reliable and AI-generated applications ready for production. He also created SlopBench, an open-source benchmark for detecting formulaic machine-generated writing.

From startup software to AI infrastructure

Cleary studied product development at New York University, graduating in 2019. While there, he co-founded Up2Code, a software-development business that became Modus after graduation and built products for early-stage startups. In 2021, he served as acting chief technology officer at Kidfund before its acquisition by UNest. He subsequently co-founded the website-feedback product Tether.

Cleary co-founded PromptHub in 2023 after his team encountered difficulties testing, versioning, and collaboratively maintaining prompts while integrating language models into software. The platform treats model instructions as managed product infrastructure: teams can compare outputs, track changes, run evaluations, and deploy updates. Its automated evaluation pipelines trigger checks from commits, reviews, and API events to catch regressions and exposed secrets before release.

Through PromptLab, Cleary also teaches model selection, agent prompting, security, shared prompt libraries, and production deployment.

• Prompting before complexity. Cleary advises teams to refine task descriptions and test prompt changes before introducing retrieval systems or elaborate agents. His early prompting work applies outside research on multiple simulated perspectives, source-grounded instructions, and emotionally framed prompts to practical experimentation.

• Provider-specific meta-prompting. PromptHub uses models to generate and refine instructions tailored to different model providers. Cleary distinguishes conventional systems from reasoning models: concise examples and explicit reasoning can improve some tasks, while unnecessary context, excessive examples, or prescribed reasoning steps can impair models with built-in reasoning capabilities. His AI Engineer session on model-sensitive prompting develops these distinctions.

• Legible agent execution. In his analysis of agent prompting, Cleary emphasizes explicit tool schemas, contextual memory, execution constraints, and separation between planning and action. He analyzes Cline and Bolt as existing examples without claiming to have created them.

• SlopBench. Cleary built an open-source benchmark for formulaic AI writing that identifies reflexive praise, canned transitions, filler, and excessive hedging through deterministic pattern matching. His account of the benchmark frames it as an opinionated measure of qualities conventional benchmarks often miss.

Applications built to ship

Founded in 2025, Converge generates full-stack applications through conversation, incorporating databases, backend logic, authentication, and deployment-ready infrastructure.

Its Convex-backed application architecture expresses schemas, permissions, business logic, and data access in TypeScript, giving coding agents more of the application’s behavior to inspect and modify directly. Cleary has described Converge’s technical direction while confirming that PromptHub continues operating alongside the newer venture.