WorkBuddy Rethinks Code Review for the AI Era

The New Office Trifecta

For decades, the office ran on Word, Excel, and PowerPoint. But as AI tools churn out Markdown, HTML, and CSV files, a new kind of document is sneaking into our workflows. WorkBuddy, an AI-powered office suite, recently updated its repository to treat these AI-native formats as first-class citizens—ones that can be edited, reviewed, and shared just like a Word doc.

This shift matters for code review. Not the kind where you stare at a pull request on GitHub, but the kind where you're reviewing an HTML page that an agent just generated, or a Markdown spec that needs human sign-off. WorkBuddy is blurring the line between document editing and code review, making it feel less like debugging and more like collaboration.

HTML: From Output to Living Document

When Claude Code engineer Thariq argued that HTML is a better output format than Markdown for AI, it sparked debate. Andrej Karpathy chimed in, suggesting that humans express needs via voice, while AI should present complex info visually. HTML, it turns out, is the bridge.

But generating HTML is only half the battle. The real challenge is editing, collaborating, and publishing it without touching the source. WorkBuddy's repository lets you do just that. You can select a block of HTML, tell the AI to tweak the background color, add a note, or rephrase a line—all without opening a code editor.

For non-programmers, this is huge. Previously, changing a webpage meant regenerating it or diving into code. Now it's like editing a document. You click, you type, you're done.

Reviewing HTML Like a Document

I ran a quick test. I fed WorkBuddy a set of project data—five initiatives, each with a lead, department, status, budget, and milestones. It turned that into a structured table, then generated an HTML dashboard that looked like a mini project management tool.

The real magic came next. I linked that dashboard to a CSV file and told the AI to pull data dynamically. After a refresh, the dashboard updated itself. No regeneration, no copy-paste. The HTML became a front-end for live data, a pattern that echoes classic MVC architecture but without the boilerplate.

For code review, this is a game-changer. You're not reviewing static code; you're reviewing a living interface. The review process becomes visual and interactive.

Publishing and Commenting: The Missing Link

One of the biggest hurdles with AI-generated HTML is sharing it. Deploying to a server, managing a domain—that's overkill for a weekly report. WorkBuddy lets you publish directly from the repository. You share a link, and colleagues can open it on their phones or in WeChat, no downloads required.

Even better, they can comment on the page itself—highlight a section, leave a note, just like annotating a PDF. This turns the HTML page into a collaborative artifact, not just a static output.

Markdown: The Unsung Hero of Review

While HTML grabs attention, WorkBuddy also upgraded smart Markdown with an AI-native review mode. When your team drafts a weekly report or a PRD in Markdown, AI suggestions appear as revisions. You approve them one by one before they merge into the final text. Once it's done, you can convert the Markdown to HTML for a polished presentation.

This workflow mirrors the classic review cycle: draft, comment, revise, approve. But now an AI is the co-author, and the review is granular and transparent.

CSV: The Data Layer That Makes It All Work

Underneath all this is CSV. WorkBuddy treats CSV as a lightweight database. I tested it with a bakery's inventory—product names, SKUs, stock levels, reorder points. The AI generated a dashboard that flagged low-stock items and allowed filtering by category. The CSV remained the source of truth, so the dashboard was always in sync.

This is where the repository becomes more than a file folder. It stores data, and that data feeds the next AI task. For instance, a team called Sunbei Bakery stores monthly cost reports in the repository, then uses them to generate the next month's analysis. The AI picks up where it left off, and the cycle continues.

The Human-Agent Review Workflow

WorkBuddy's repository isn't just for individuals; it has team spaces with permissions. You can invite colleagues to view, edit, and comment, and agents can participate too. This means a review can involve multiple humans and multiple agents, all working on the same file.

In my tests, the experience felt closer to a real office tool than a tech demo. I could select a risk module in the HTML, ask the AI to lighten the background and add a note: "Need a project meeting by Friday to resolve." The change applied only to that selection, leaving the rest intact.

Beyond the Office: Personal Use Cases

The repository also handles personal projects. I created a "life battery" dashboard from a CSV of daily energy levels, and a couple could track household chores and "redemption items" in a shared ledger. You could even build a lunch roulette system based on nearby restaurants' prices and ratings.

This flexibility is what makes the repository feel less like a feature and more like a new way to work. It's not replacing professional databases or SaaS tools, but for personal and small-team needs, it's enough.

A New Kind of File for a New Kind of Work

WorkBuddy isn't killing Word, Excel, or PowerPoint. It's still supporting them, and its "human-AI co-writing" feature already lets agents edit those formats. But it's also embracing the formats AI produces naturally.

So now we have two generations of files: the traditional Office trio, built for human input via mouse and keyboard, and the AI-native trio of Markdown, HTML, and CSV, built for agents to generate, understand, and modify.

As AI becomes a regular participant in writing, editing, and data processing, the file we open each morning is no longer a static artifact. It's a living thing that can be updated, reviewed, and shared—by humans and agents alike.

In the end, file formats are more than just extensions. They define how we produce, modify, and hand off information. The participants have expanded from human-only to include agents, and WorkBuddy is showing how code review fits into that new reality.