Goose with Rizel Scarlett and Ebony Louis

Anthony Campolo joins Block’s Rizel Scarlett and Ebony Louis to discuss Goose, the open‑source AI agent that speeds up coding, content creation, and daily tasks.

Episode Description

Anthony Campolo joins Block’s Rizel Scarlett and Ebony Louis to discuss Goose, the open‑source AI agent that speeds up coding, content creation, and daily tasks.

Episode Summary

In this wide‑ranging 69‑minute conversation, Anthony learns how Goose grew from an internal experiment at Block into an open‑source developer agent that combines large‑language‑model reasoning with MCP‑based tool calling. Rizel and Ebony recount their unconventional paths from coding bootcamp to Developer Advocacy, explain how Goose migrated codebases, automates repetitive chores, schedules calendar events, and even controls local hardware. Live demos showcase Goose’s desktop and CLI interfaces, model switching between Claude and GPT‑4, automated game generation, image‑creation hiccups, and practical tricks such as killing stray ports or summarizing long chats to stay within context limits. The trio also covers sharing recipes, running agents on cron jobs, navigating the growing MCP server registry, and pairing Goose with voice tools like ElevenLabs. Throughout, they reflect on AI’s impact on junior engineers, documentation workflows, conference life, and the importance of trust and transparency when giving software real autonomy.

Chapters

00:00:00 - Introductions and Career Journeys

Anthony welcomes returning guest Rizel Scarlett and first‑timer Ebony Louis, inviting each to share how a shared coding‑bootcamp background ultimately led them into Developer Advocacy at Block. They compare early software roles, talk through Ebony’s pivot from speech‑language pathology to tech, and set the stage for exploring Block’s growing family of brands.

The hosts outline the plan for the session—an informal tour of Goose—while noting the audience’s live‑stream questions and Anthony’s curiosity about practical AI tooling. This personal framing helps listeners understand each speaker’s vantage point before the technical deep‑dive that follows.

00:05:05 - From Square to Block: Clearing Up the Branding

Confusion over Block’s identity gives Rizel a chance to unpack the 2021 rebrand from Square and list sister products like Cash App, Tidal, and Afterpay. Anthony recalls Jack Dorsey’s early crypto advocacy and jokes about rumors linking him to Satoshi Nakamoto, illustrating Block’s deep Web3 roots.

Recognizing how a “company of companies” structure affects developer tooling, the guests explain why an umbrella brand made sense and how it fosters cross‑vertical innovation. This sets a natural bridge to Goose, which began as an internal tool shared across multiple Block teams.

00:10:15 - Why Goose Exists: An AI Agent for Everyone

Ebony recounts Goose’s origin as a small script that automated large‑scale code migrations, gradually expanding into a versatile, open‑source agent once other teams felt its productivity boost. Rizel emphasizes that the agent excels at both engineering and non‑technical workflows, making it approachable for newcomers.

They discuss Goose’s MCP foundation, its ability to plug into any compliant server, and the decision to share it publicly. Listeners gain context for why a fintech giant would invest in a general developer agent and how Goose’s open architecture accelerates experimentation.

00:15:12 - Touring the Goose Interface

Rizel launches the desktop app, highlighting its Warp‑like terminal aesthetic, quick‑start prompts, and provider picker that spans GPT‑4, Claude 3.5/3.7/4, and in‑house models. She contrasts Goose’s GUI with the CLI, noting why each resonates with different types of developers.

Live switching between models sparks a discussion of subjective “feel” versus objective accuracy, and the team shows how Goose exposes version toggles so users can benchmark responses on the fly. The demo grounds abstract agent concepts in a tangible UI.

00:20:18 - Building a Game and Comparing Models

A simple “make a Snake game in Python” prompt lets Goose scaffold files, install dependencies, and run tests autonomously. Anthony is impressed that Claude 4 not only writes code but executes and verifies it, while Claude 3.5 stops short.

This side‑by‑side test underscores why model choice matters and how Goose’s context meter helps users avoid hitting token limits mid‑project. The segment closes with tips for embedding consistent coding conventions into session instructions.

00:25:25 - Image Generation Experiments

Pivoting to creative work, the trio asks Goose to design a YouTube thumbnail for the stream. They critique the first draft’s cut‑off text, prompting a second iteration focused on aspect ratio and legibility. The shortcomings spark a candid chat about the current gap between text‑to‑image dreaminess and precise brand‑asset production.

Rizel admits she seldom uses AI art outside of quick resizing jobs, while Anthony notes that newer multimodal models are narrowing the fidelity gap. The hands‑on failure is as instructive as a success, revealing real‑world limits.

00:30:30 - Prompt Iteration and Model Limitations

Multiple retries highlight common image‑model pitfalls—ignored design constraints, uncanny visuals, and inconsistent adherence to instructions. The hosts share prompt‑engineering hacks, such as reinforcing specification clauses and requesting iterative refinement instead of full redraws.

They broaden the lens to discuss why many image models were trained to prioritize rich scenes over tight layout control and how recent API updates aim to balance creativity with compliance. The conversation illustrates continuous, collaborative improvement rather than magic.

00:35:35 - AI in Daily Workflows: A Junior Engineer’s View

Ebony explains how AI replaced hours of Stack Overflow searches during her first engineering roles, speeding up debugging, brainstorming, and content ideation. She argues that newcomers still need fundamentals but can advance faster when rote roadblocks shrink.

Anthony echoes the sentiment, sharing personal stories of treating ChatGPT as a sounding board for history research and recipe planning. Their reflections normalize AI augmentation without hyperbole, stressing that curiosity and judgment remain essential.

00:40:40 - Calendar Hacks, Open Ports, and Automation Comfort Zones

A live attempt to schedule a meeting shows Goose’s Google‑Calendar integration—and its timezone quirks. Anthony contrasts Google‑centric workflows with his privacy‑focused Notion setup, sparking thoughts on OAuth flows and multi‑provider support.

Rizel demonstrates killing orphaned Dev Server ports via a single Goose instruction, then discovers the “Open Ports” macOS utility Anthony recommends. The back‑and‑forth demystifies agent‑driven system commands while acknowledging when human oversight still feels safer.

00:45:45 - Managing Context, Memory, and Summaries

Long chats risk exceeding model limits, so Rizel reveals Goose’s context percentage meter and one‑click summarizer that compresses history without losing thread coherence. They contrast manual restarts, memory pins, and autonomous pruning, weighing clarity against conversational flow.

Listeners learn practical tactics—store key facts in memory, start fresh sessions when logic drifts, and use summarization before heavy tool‑calling—to keep agents sharp over multi‑hour tasks.

Goose “recipes” package a session’s tools, prompts, and memories into shareable agents. Ebony describes adding onboarding bubbles so colleagues know where to begin, while Rizel previews scheduled runs and cron‑style triggers coming in the next release.

They outline four autonomy modes—from chat‑only to fully autonomous—helping skeptics adjust trust gradually. The segment clarifies that responsible agent design pairs power with guardrails and auditability.

00:55:55 - Exploring the MCP Ecosystem

Discussion shifts to the wider MCP spec that lets Goose talk to anything from Blue‑Sky feeds to GitHub issues. Anthony praises the Glamour AI registry, where servers earn safety report cards before adoption.

Rizel and Ebony recount Block’s contributions to the spec and their migration of Goose’s client from Python to Rust for performance. Resources for learning MCP, including visual guides and YouTube tutorials, round out the technical tour.

01:01:00 - Voice Generation and Multimedia Workflows

ElevenLabs integration inspires talk of AI‑narrated blog posts, audiobooks, and conference demos where agents speak with expressive intonation. Ebony’s short video that lets Goose “take over the script” illustrates playful marketing without sacrificing clarity.

Anthony shares his own pipeline—deep‑research reports turned into narrated podcasts—and muses on multimodal agents that write, design, and voice content end‑to‑end. The hosts agree that ethical guardrails are vital as synthetic voices approach realism.

01:05:30 - DevRel Life, Conferences, and Closing Thoughts

The conversation winds down with a peek at Block’s expanding DevRel team, the realities of newborn‑friendly conference schedules, and Midwest summer heat at KCDC. Anthony’s local tornado sirens add real‑time drama, yet camaraderie remains lighthearted.

Rizel and Ebony recap key takeaways—download Goose, explore agents safely, and embrace AI as a collaborative partner. They share social handles before Anthony promises an Auto‑Show summary of the episode, ending at 01:09:15 with gratitude to the live chat and an open invitation to experiment.