Rizel Scarlett

Agents fundamentally changed how we program. They enable developers to move faster by disintermediating the traditional development workflow. This means less time switching between specialized tools and fewer dependencies on other teams. Now that agents can execute complicated tasks, developers face a new challenge: using them effectively over long sessions.

The biggest challenge is context rot. Because agents have limited memory, a session that runs too long can cause them to "forget" earlier instructions. This leads to unreliable outputs, frustration, and subtle but grave mistakes in your codebase. One promising solution is Code Mode.

Everyone on Tech Twitter has been buying Mac Minis, so they could run a local agentic tool called OpenClaw. OpenClaw is a messaging-based AI assistant that connects to platforms such as Discord and Telegram allowing you to interact with an AI agent through DMs or @mentions. Under the hood, it uses an agent called Pi to execute tasks, browse the web, write code, and more.

Seeing the hype made me want to get my hands dirty. I wanted to see if I could build a lite version for myself. I wanted something minimal that used goose as the engine instead of Pi. I tentatively dubbed it AltOpenClaw.

MCP Apps allow you to render interactive UI directly inside any agent supporting the Model Context Protocol. Instead of a wall of text, your agent can now provide a functional chart, a checkout form, or a video player. This bridges the gap in agentic workflows: clicking a button is often clearer than describing the action you hope an agent executes.

MCP Apps originated as MCP-UI, an experimental project. After adoption by early clients like goose, the MCP maintainers incorporated it as an official extension. Today, it's supported by clients like goose, MCPJam, Claude, ChatGPT, and Postman.

Even though MCP Apps use web technologies, building one isn't the same as building a traditional web app. Your UI runs inside an agent you don't control, communicates with a model that can't see user interactions, and needs to feel native across multiple hosts.

After implementing MCP App support in our own hosts and building several individual apps to run on them, here are the practical lessons we've picked up along the way.

Can you automate taste? The short answer is no, you cannot automate taste, but I did make my design preferences legible.

But for those interested in my experiment, I'll share the longer answer: I wanted to participate in Genuary, the annual challenge where people create one piece of creative coding every day in January.

My goal here wasn't to "outsource" my creativity. Instead, I wanted to use Genuary as a sandbox to learn agentic engineering workflows. These workflows are becoming the standard for how developers work with technology. To keep my skills sharp, I used goose to experiment with these workflows in small, daily bursts.

As AI agents grow in capability, more people feel empowered to code and contribute to open source. The ceiling feels higher than ever. That is a net positive for the ecosystem, but it also changes the day-to-day reality for maintainers. Maintainers like the goose team face a growing volume of pull requests and issues, often faster than they can realistically process.

We embraced this reality and put goose to work on its own backlog.

One day, we will tell our kids we used to have to wait for agents, but they won't know that world because the agents in their day would be so fast. I joked about this with Nick Cooper, an MCP Steering Committee Member from OpenAI, and Bradley Axen, the creator of goose. They both chuckled at the thought because they understand exactly how clunky and experimental our current "dial-up era" of agentic workflows can feel.

Model Context Protocol (MCP) has moved the needle by introducing a new norm: the ability to connect agents to everyday apps. However, the experience isn't perfect. We are still figuring out how to balance the power of these tools with the technical constraints of the models themselves.

To plan or not to plan, that's the wrong question. Rather than a binary yes/no, planning exists on a spectrum. The real question is which approach fits your current task and working style.

Different developers approach planning in different ways. One builder might draft detailed pseudocode before touching a keyboard, while another practices test driven development to let the architecture emerge organically. You'll find teams sketching complex diagrams on whiteboards and others spinning up fast prototypes to "fail fast" and refactor later.

If planning is a spectrum when coding manually, why wouldn't it be a spectrum when using an agent to code as well?

AI agents are often described as brilliant, overeager interns. They're desperate to help, but sometimes that enthusiasm leads to changes you never asked for. This is by design: the large language models powering agents are trained to be helpful. But in code, unchecked helpfulness can create chaos. Even with clear instructions and a meticulous plan, you might hear, "Let me just change this too…" A modification that's either unnecessary or, worse, never surfaced for review.

Sure, you can scour git diff to find and revert issues. But in a multi-step process touching dozens of files, untangling one small, unwanted change becomes a manual nightmare. I've spent hours combing through 70 files to undo a single "helpful" adjustment. Asking the agent to revert is often futile, as conversational memory isn't a snapshot of your codebase.

You've heard the buzz: AI is reshaping our work. Maybe you've tinkered with ChatGPT, or your company is pushing you to "level up." But between the hype and the endless tutorials, a gnawing question remains: how do you move from theory to building something real?

The answer is practice. Not just following steps, but creating, problem-solving, and learning by doing.

That's why we're launching Advent of AI, a 17-day challenge series starting December 1st. Whether you're a beginner taking your first steps or an advanced developer exploring AI agents, this is for you. Each weekday, you'll get a new, hands-on project designed to transform you from an AI spectator into a confident builder.

"Migrate my app from x language to y language." You hit enter, watch your AI agent spin its wheels, and eventually every success story you've heard feels like a carefully orchestrated lie.

Most failures have less to do with the agent's capability and more to do with poor prompt and context strategy. Think about it: if someone dropped you into a complex, unfamiliar codebase and said "migrate this," you'd be lost without a plan. You'd need to explore the code, ask questions about its structure, and break the work into manageable steps.

Your AI agent needs the same approach: guided exploration, strategic questions, and decomposed tasks.