About Grace Fletcher

Overview

I build the systems, workflows, and technical knowledge organizations depend on to enable developers, and increasingly AI systems, to adopt complex technical products.

Over more than 13 years, that has meant founding documentation functions, designing developer portals, restructuring information architectures, devising custom API tooling, and leading cross-functional AI-readiness work.

The implementations looked different at TUNE, Branch, Foursquare, and Chainlink Labs, but they repeatedly addressed the same organizational challenge: technical knowledge no longer matched the products, audiences, or operating conditions it needed to support.

My professional work focuses on redesigning those systems. The writing, conceptual models, and diagnostics on this site make the recurring patterns behind that work visible and useful beyond the organizations where I first encountered them.

What I have built

Across four organizations, I have repeatedly been brought in when technical knowledge had become strategically important but the systems supporting it were incomplete, fragmented, or no longer fit for purpose.

How I got here

My path into this work began in academic research rather than industry.

My academic background is in comparative and developmental cognition, the study of how different minds represent and reason about the world. That training shaped how I approach evidence, interpretation, hidden assumptions, and the gap between what a system appears to measure and what it actually captures.

I later moved into technical communication and developer experience. The most consequential problems were rarely about writing quality alone. They concerned what technical knowledge was for, who needed to use it, where judgment lived, and whether its structure still matched the environment around it.

At Chainlink, that question became concrete in a new way. As AI agents became consumers of developer documentation, assumptions that had long supported human readers stopped holding. People could infer missing context, recognize what applied, and recover from ambiguity. AI systems could not be expected to do so reliably.

The documentation had not simply become worse. The consumer had changed. Redesigning the information architecture for that new consumer improved an independent agent-readiness assessment from 65% to 92%.

That work crystallized the territory I now focus on: what organizations must redesign when AI changes how technical knowledge is produced, retrieved, evaluated, and used.

Where this work becomes useful

My work is most useful when an organization needs to:

• Establish technical knowledge as an organizational capability. This includes defining ownership, workflows, standards, roadmaps, governance, and the teams responsible for maintaining the system.
• Turn fragmented expertise into a structure that can scale. Critical knowledge often exists across source code, engineering teams, product decisions, support channels, and undocumented practice. I make that knowledge usable beyond the people who currently hold it.
• Align information architecture with how people adopt a product. I redesign documentation and developer experiences around user intent, implementation paths, product relationships, and the decisions users actually need to make.
• Adapt technical knowledge for AI-assisted production and consumption. This includes retrieval-oriented architecture, machine-consumable content, AI-assisted authoring, validation workflows, and structures that keep generated output reviewable rather than merely faster.

Current practice and tools

I use AI throughout the work itself, including research, synthesis, restructuring, authoring, coding, evaluation, and critique. Tool familiarity matters, but only in combination with the architecture, judgment, and validation needed to produce reliable outcomes.

I do not treat a list of tools as evidence of capability. The important question is whether the resulting system helps people and AI understand, implement, and evaluate complex technical products more reliably.

How the site connects to the work

The conceptual work on this site begins with patterns observed in implementation.

A documentation system may appear accurate but still fail because a new consumer cannot recover the context human readers supplied automatically. AI-assisted production may increase output while making validation, judgment, and accountability harder to locate. A familiar signal of expertise may remain persuasive after it stops reliably representing the capability an organization intended to evaluate.

The writing identifies these changes. The conceptual models explain the mechanisms producing them. The diagnostics help organizations examine whether the same conditions are present in their own systems.

That progression, from implementation to recurring pattern to practical application, connects my professional work with the body of thought developed here.

Why this work matters differently now

AI has not created every knowledge, documentation, or expertise problem organizations now face. It has changed the conditions under which those problems can remain hidden.

Generation can scale faster than validation. Plausibility can influence evaluation before verification is complete. Technical information can be accurate but still fail when an AI consumer cannot recover the context a person would supply automatically.

The challenge is not simply to produce more information or adopt more AI tools. It is to redesign the systems that preserve context, locate judgment, transfer expertise, and make responsibility visible.