Summary of Developers who move fast still need to do it together

Overview of Developers who move fast still need to do it together

In this MS Build conversation, Ryan Donovan speaks with Cassidy Williams, Senior Director of Developer Advocacy at GitHub, about the rapidly changing world of AI-assisted software development. The discussion centers on GitHub’s new Copilot app, agentic workflows, local and BYOK model options, and the growing need for developer judgment, community, and collaboration even as tools become more automated.

Key Announcements and Product Direction

GitHub Copilot app and agentic workflows

Cassidy highlights GitHub’s new Copilot app as a central interface for working with AI agents. It builds on the Copilot CLI and a new Rust SDK, aiming to make agent interaction faster and more performant.

Key capabilities discussed:

• Chat-based interaction with agents
• Automations and task execution
• A new canvas concept for interacting with agents through a UI rather than only text
• Approval-style gestures, such as thumbs up/down for commits

Foundry and model flexibility

GitHub is also leaning into flexibility with Foundry, which supports:

• Running local models
• Using open-source models
• Bringing your own API key

The broader message: developers should be able to choose the models and workflows that fit their needs, while GitHub provides the tooling around them.

How AI Is Changing the Developer Workflow

More planning, less hands-on coding

Cassidy says her own process has shifted significantly:

• She now spends more time on planning, specifications, and decision-making
• Less time is spent on writing code line-by-line
• Many of the decisions that used to happen while coding now happen earlier, during specification

Decision fatigue and cognitive overload

A major theme is that AI tools may not reduce work so much as compress it:

• Developers are doing more in a shorter amount of time
• Instead of a few PRs a week, teams may now produce many more changes per day
• This creates cognitive overload and “decision fatigue”

GitHub is trying to address this by making agent workflows more visual and easier to manage.

The Role of UI, Canvas, and Multitasking

Cassidy is particularly excited about the idea of a canvas for agents:

• Developers can create custom interfaces to manage agent work
• One example: turning issues into a Kanban board inside the app
• This helps separate tasks mentally and reduces the chaos of juggling many sessions

She argues that developers have leaned heavily into CLI and headless tools, but a good UI overlay can improve clarity and multitasking.

Human Judgment Still Matters

Community and peer review are more valuable than ever

A strong takeaway from the interview is that human feedback remains critical:

• Cassidy relies on detailed specs
• She also shares them with developer group chats for candid feedback
• Human reviewers are much more blunt and useful than AI, which tends to be sycophantic and overly approving

Developers still need taste, creativity, and opinion

Cassidy pushes back on the idea that AI removes the need for developers:

• Developers still provide taste and judgment
• Code may be changing, but software engineering is not disappearing
• The job is shifting toward steering, reviewing, and specifying rather than simply typing code

Cost, Pricing Pressure, and Local Models

Real usage pricing is changing behavior

The conversation touches on pricing shocks tied to AI usage:

• GitHub is moving toward pass-through pricing from model providers
• This has created surprise for some customers
• GitHub was relatively late to this model because it wanted to avoid hurting developers too much

Cost controls and optimization tools

GitHub is responding with tools like:

• Bring Your Own Key
• Local model support
• Aggregators and partners such as OpenRouter and Fireworks
• CLI cost guidance, including a chronicle cost tips command to suggest cheaper or more efficient usage patterns

The result is a more configurable but also more cost-conscious AI development environment.

Developer Identity and Culture Shifts

A changing sense of what it means to be a developer

Cassidy notes that many developers are experiencing an identity crisis:

• Some still see coding as an artisanal craft
• Others embrace AI-generated output
• Many fall somewhere in between

She compares the shift to power tools in carpentry: the role changes, but the profession doesn’t disappear.

More analog habits may return

As developers deal with more automation and more cognitive load, Cassidy expects:

• More handwriting
• More analog workflows
• More creative and personal projects

She sees AI lowering the barrier to building, which may lead to more weird, expressive, and human-made things.

Broader Outlook on the Future of Development

“It depends” at scale

A recurring idea is that AI development is highly context-dependent:

• Different codebases need different approaches
• Sometimes AI is enough
• Sometimes code should still be written manually for precision
• The right workflow depends on the task, context, and tolerance for error

The future is still unsettled

Cassidy is optimistic but cautious:

• The pace of change is too fast to predict with certainty
• More AI tooling is likely
• Some AI hype may settle into more practical SaaS and infrastructure patterns
• The real frontier is finding the right balance between automation, control, and human judgment

Notable Takeaways

• AI tools are shifting developers from writing code to writing specs and guiding systems.
• Human collaboration is becoming more important, not less.
• GitHub is betting on flexibility: local models, BYOK, open integration, and UI-based agent control.
• The biggest challenge ahead may be managing overload, cost, and developer identity—not just building faster.
• Despite all the automation, Cassidy argues there is still a strong need for developers, taste, and community.

Closing Thought

The episode’s core message is that AI is making developers faster, but not necessarily lighter on work. The future of development looks less like replacing humans and more like helping humans coordinate, decide, and create together—at higher speed and with better tools.