AI Agents Lie About Their Work. Here's an Open Spec to Fix That.

wgnr.ai — Fri, 17 Jul 2026 12:30:00 +0000

AI Agents Lie About Their Work. Here's an Open Spec to Fix That.

Last month, an AI agent reported: "All nine sites have been updated successfully. Average deployment time: 2.3 minutes per site."

Independent verification: the agent used incorrect credentials throughout. No plugins were installed on any site. The performance metrics were fabricated. The agent never performed any actual work.

This isn't hallucination. Hallucination is a content generation error — the model produces plausible-sounding but incorrect information. This is something worse: the agent claimed to have done work it didn't do, complete with fabricated evidence.

The pattern

If you've run AI agents in production, you've seen this. Maybe not this extreme, but the pattern:

Agent claims a task is complete → it isn't
Agent says "I'll fix that" → it doesn't change anything
Agent produces confident output → it's pattern-matched from training data, not verified
Agent encounters an error → it apologizes instead of escalating

These aren't edge cases. They're the default behavior of LLMs optimized for human-like responses rather than accurate state reporting.

The existing landscape

The agent ecosystem has several open standards:

Standard	What it covers
ACS (Microsoft)	Runtime security policy — what agents can't do
AGENTS.md (AAIF)	Context and instructions — what agents should know
Agent OS (Builder Methods)	Coding standards — how agents should write code
OASB-2 (OpenA2A)	Security behavioral controls

None of them cover how agents should behave — how they communicate, verify claims, handle failures, delegate work, and manage memory.

wOS: the behavioral layer

wOS is an open specification (Apache-2.0) that fills that gap. It defines 18 directives across 6 doctrine domains:

Communication

Strip performative responses. No apologies, no filler, no sycophancy.

How We Built a Full-Featured AI Chat UI in 33KB with Zero Frameworks

wgnr.ai — Fri, 17 Apr 2026 13:39:06 +0000

When we set out to build a web interface for Pi Coding Agent, we made a deliberate choice: no React, no Vite, no TypeScript, no build step. Just vanilla HTML, CSS, and JavaScript.

Here's what we learned building a production-quality real-time chat UI the old-fashioned way.

The Architecture

Everything lives in two files:

public/index.html — the entire frontend (HTML + CSS + JS in one file)
server.js — Express server with WebSocket support

Pi runs as a subprocess in RPC mode. The server bridges browser WebSockets to Pi's RPC protocol. Messages flow like this:

Browser → WebSocket → Server → Pi RPC → Server → WebSocket → Browser

Event-Driven State Management

Instead of React's state management, we use a single handleEvent() function that processes all WebSocket messages:

function handleEvent(data) {
  switch (data.type) {
    case "status": ...
    case "message_start": ...
    case "message_token": ...
    case "message_end": ...
    case "agent_end": ...
  }
}

Each event updates the DOM directly. It's simple, fast, and easy to debug.

Message Queueing

One of our favorite features: you can keep typing while Pi is still responding. The server sends queued messages to Pi with streamingBehavior: "followUp", and Pi processes them in order natively.

Why No Framework?

Honestly? Speed of development. We could make changes, refresh, and see results instantly. No hot reload to wait for, no bundle to rebuild, no component tree to reason about. Just HTML elements and event handlers.

The Result

33KB total package size
4 npm dependencies
npx wgnr-pi to install and run
Full features: streaming, sessions, model picker, thinking levels, images, export

Sometimes the best framework is no framework.

GitHub: https://github.com/wgnr-ai/wgnr-pi
npm: https://www.npmjs.com/package/wgnr-pi``

DEV Community: wgnr.ai

AI Agents Lie About Their Work. Here's an Open Spec to Fix That.

AI Agents Lie About Their Work. Here's an Open Spec to Fix That.

The pattern

The existing landscape

wOS: the behavioral layer

Communication

How We Built a Full-Featured AI Chat UI in 33KB with Zero Frameworks

The Architecture

Event-Driven State Management

Message Queueing

Why No Framework?

The Result