Hardware FixRecommendedDevice not working? Your driver may be the problemCheck updates for common hardware issues.Fix DriversOctober DealsAmazon USOctober deal check: compare before you payAmazon US: current deals, useful picks and tech finds.Check DealsSlow PC?RecommendedPC slow today? Run a repair scan before it gets worseResolve common Windows issues and optimize system performance.Scan Now×
Skip to content
World desk4 min

Why I Stopped Handing Agents a Framework

Jonas Gauffin argues that explicit, searchable frontend code can make agent-written changes easier to trace and review in small-to-medium SPAs, while acknowledging when frameworks still fit better.
Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Jonas Gauffin’s case for writing explicit frontend code instead of starting with Vue or Angular is not that frameworks are bad, or that coding agents cannot use them. It is that for his small-to-medium single-page applications, he finds code easier for an agent to reason about when the behavior is visible in the files, searchable by name, and straightforward to review. He makes that case with an important condition: a person reviews the generated changes.

Why framework behavior can be hard for an agent to follow

Gauffin describes an agent’s practical feedback loop as reading files, searching for names, running type checks, and running tests. In his experience, that loop can leave gaps when the behavior depends on runtime mechanisms such as scheduling, reactive dependencies, or change detection. The agent may have to infer what the application will do from code whose effects are not immediately visible in the files it can inspect.

His response is to favor direct updates and explicit connections, so the cause of a behavior is easier to locate and a change is easier to inspect. That is an engineering argument based on his own work with Vue, Angular, and his library @relax.js/core—not a controlled comparison or a claim that framework code is inherently unsuitable for agents.

The three qualities he wants from agent-written code

Failure locality

When a symptom appears, the likely cause should be close to the code showing it. Gauffin argues this is harder when the explanation may be distributed across a scheduler, dependency graph, or zone rather than apparent in the affected file.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Greppability

Events and connections should have names that can be searched across their producers and consumers. Searching for a name gives an agent a concrete route through the code instead of requiring it to reconstruct an implicit relationship.

Reviewability

A diff should expose the intended behavior clearly enough for a human—or an agent—to check it. This makes review a design criterion, not merely a final quality-control step.

These are Gauffin’s criteria for making code workable in his agent-assisted process. They are not measured performance results: the essay reports no benchmark, sample size, or productivity figure.

Explicit code is not enough on its own

Gauffin says the workflow also depends on support that helps an agent avoid predictable errors and verify what it writes. He describes three pieces in @relax.js/core:

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
  • Short skills: guidance files loaded before work begins, aimed at patterns he says agents commonly get wrong. The detailed documentation remains the reference for APIs and mechanisms.
  • Visible errors: a template path that fails to resolve might otherwise render as an empty string. In his account, the library sends such failures through an error channel, and a test helper turns that channel into assertions.
  • A template checker: he says npx @relax.js/core check resolves template expressions against TypeScript types at the call site and reports compiler-style messages. Gauffin says this addresses much of the template-checking gap he sees with Angular without adding a compiler to the build; that is his description of his tool, not an independently verified comparison.

He also names mount(), flush(), fakeServer(), and mountRouting() as test seams used with Vitest. The purpose is to let an agent check behavior through tests rather than depend on a person manually clicking through the interface.

How skills and documentation divide the work

In a follow-up, Gauffin distinguishes agent guidance from reference material: skills are meant to correct a wrong default and direct the agent to deeper documentation; docs explain APIs and how mechanisms work. He sums up the distinction this way: “Would an agent that never read this produce code that compiles, type-checks and does nothing? Skill. Would it merely not know a name? Docs.” This is his proposed rule, not a general standard.

He says npx @relax.js/core init-agents writes seven skill files covering areas including the core model, templates, forms, routing, services, testing, and setup. These setup details describe the author’s follow-up and should not be read as independent verification of the package.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

When Vue or Angular may still be the better choice

Gauffin explicitly identifies situations where he would keep an established framework:

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
  • Agent familiarity matters most: if getting a strong first draft without loading skills is the priority, an agent’s existing familiarity with a framework may be more valuable than the explicit-code approach he favors.
  • State is deeply interdependent: applications with complex relationships between state may be a better fit for a framework.
  • Server-side rendering is required: he lists SSR as another reason to choose Vue or Angular rather than his proposed approach.

His suggested fit is narrower: a small-to-medium SPA where skills and automated checks support the agent, and a human reviews the diff. The choice therefore depends not only on code style but also on the application’s state model, rendering needs, and willingness to maintain agent-specific guidance.

What this argument does—and does not—establish

The essay is a first-person engineering case for making behavior easier for coding agents to inspect. It does not establish that Vue or Angular produces worse results, that agents cannot work with those frameworks, or that @relax.js/core is more productive. Gauffin’s examples are qualitative, and his checker and testing workflow are described by their author rather than independently assessed.

The practical takeaway is to evaluate the whole workflow: what runtime behavior an agent must infer, how easily it can search for connections, whether tests expose failures clearly, and whether a person can review the resulting changes. For Gauffin’s stated use case, explicit code plus skills, actionable errors, a checker, and test seams makes that workflow manageable; the same reasons do not automatically apply to every frontend project.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Leave a Reply

Your email address will not be published. Required fields are marked *

What’s actually slowing this PC down?

Pick the symptom - the matching free tool is one click away.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

More from the Wire

  1. World desk4 min
    How to Spot an AI Voice Scam Before Sending MoneyDon’t rely on how a caller sounds. Pause, call back through a known number, and verify the emergency with another trusted person before sending money.
  2. Mountain View desk4 min
    Google’s SynthID Detector: How to Check AI-Generated Images, Video and AudioGoogle’s SynthID Detector looks for an embedded watermark in supported images, video and audio. Here is what its results do—and do not—show.
  3. Redmond desk20 min
    How to create a link to File or Folder in Windows 11Windows 11 gives you several ways to point to a file or folder without moving or duplicating it. You can create a desktop shortcut,…
Recommended PC Tool
Recommended PC Tool
Windows Errors? Fix Them Before They SpreadFree repair scan
Crashes, No Sound, or Screen Glitches?Free driver scan

Two free Windows tools

One Free Minute Could Fix That PC

Before you go - each of these free tools takes about a minute and tackles what quietly slows a Windows PC down.

Special offer. View Outbyte info, uninstall instructions, EULA, and Privacy Policy.