Selected Prolego research and teaching

Prolego has worked exclusively in AI since 2017. Our public teaching offers a record of the questions we investigated and the methods we developed as the technology changed. These selected contributions help prospective clients understand where our judgment comes from, alongside our enterprise delivery experience.

Our current approach has evolved through that work. Historical articles explain what we were investigating at the time; our principles and new playbook describe how we work today.

Helping business leaders see the opportunity: MIT, 2019

Kevin Dewalt gave an MIT Bootcamps e-seminar on April 29, 2019 about generating business opportunities with AI. The 58-minute session connected technical capabilities to business opportunities for a nontechnical audience.

That business perspective remains central to Prolego: help leaders understand what becomes possible and what they should build. Today, we frame that question around re-engineering business economics and AI Abundance.

Explore: official MIT record and recording.

Building reliability: evaluations in 2023 and PDD in 2024

Our August 2023 evaluation article and engineering discussion compared model and instruction combinations on correctness, failure, speed and cost. We were investigating how a team could inspect behavior and use evidence to guide engineering, rather than judge an application by a convincing demonstration.

The 2024 public Performance-Driven Development guide developed the method through representative tasks, evaluations and expert feedback. It includes a Formula 1 example and a readable performance report.

Our current PDD method centers on the two parts of reliability: organizational consensus about what the AI should do and ensuring it does it correctly. It incorporates stakeholder evaluation, production auditing and reviewed AI-assisted improvements developed through subsequent enterprise work.

Explore: evaluation article, engineering discussion, public PDD guide and evaluation report explanation.

Identifying a different engineering role: September 2023

Kevin and Justin Pounders discussed the AI Systems Engineer role on September 13, 2023. The work emphasized experimentation, systems thinking and communication when building around a capable but variable model interface. An accompanying job-description template helped companies think about whom to hire.

Our current hiring perspective emphasizes systems thinking, communication, fast learning and people already building and working with AI. Specialized model-training credentials are not the defining qualification.

Explore: article and template and engineering discussion.

Testing what drives performance: the 2024 RAG study

Justin Pounders studied a question-answering application using Formula 1 rules, examining model choice, retrieval, added definitions and agent behavior. The expanded study used 25 human-developed questions, each asked three times, with evaluation and human review. Relevant context produced the major performance improvement; smaller and larger models became more similar on this test.

This was a controlled investigation of a particular application, useful for understanding how the surrounding system affects performance. Its enduring contribution is the experimental method and the importance of supplying the information the task needs. It is not a current ranking of models or a reason to spend months selecting one.

Explore: expanded study and earlier article.

Connecting research to enterprise engineering: Pinecone, April 2024

The April 11, 2024 Pinecone-hosted workshop brought together Pinecone's research and Prolego's Formula 1 study. Kevin presented practical enterprise questions about policy, cost and speed. Justin discussed implementation and evaluation. The discussion examined document hierarchy, definitions, retrieval and whether an evaluator measures the right thing.

Explore: workshop recording and dated announcement.

From early experiments to the current playbook

After our full pivot to generative AI in March 2023, we invested our own engineering effort and worked with early adopters to understand how to deploy it inside real enterprises. The work changed our view of the dominant risk: as building became easier, agreeing on behavior and making it reliable became central. Our company history explains that transition.

The 2024 AI Abundance executive workbook connected transformation ambition to practical investment and capability development. Our current definition is sharper: AI Abundance is a future in which intelligence no longer constrains the economic growth of your business, with AI at the core of every economically consequential decision.

Enterprise delivery continues to shape the method: think big, start small, establish Operations Intelligence and expand useful capability as AI advances. Our case accounts show how that develops through actual work. Our AI-native team practices explain the emerging changes to communication, shared knowledge and decisions.

Explore: 2024 executive workbook, 2024 architecture discussion and Prolego's YouTube channel.