Enny AIEnny AI
AI Agents
F

Full-Stack Development

Production-ready applications with React, Node.js, and cloud infrastructure.

A

AI Integration

GPT-4, Claude, and custom AI pipelines built into your products.

D

Database & Backend

PostgreSQL, Prisma, and scalable data architectures.

A

Automation & Workflows

n8n, Make.com, and custom integrations that eliminate manual work.

R

Rapid Prototyping

Bubble.io and Voiceflow for fast MVPs and chatbots.

T

Technical Consulting

Architecture reviews, tech stack selection, and strategic guidance.

View all services→
ToolsWorkAboutInsightsContact
Get in TouchStart a Project
AI Agents
Full-Stack DevelopmentAI IntegrationDatabase & BackendAutomation & WorkflowsRapid PrototypingTechnical Consulting
ToolsWorkAboutInsightsContact
Get in TouchStart a Project
Enny AIEnny AI

Full-stack development, AI integration, and rapid prototyping. From custom code to no-code — whatever ships fastest.

Services

  • Full-Stack Development
  • AI Integration
  • Database & Backend
  • Automation & Workflows
  • Rapid Prototyping
  • Technical Consulting

Company

  • About Us
  • Our Work
  • Insights
  • Contact

Contact

  • derek@enny.ai
  • Truckee, California

© 2026 Enny AI. All rights reserved.

Privacy PolicyTerms of Service
All Tools
ICLR 2026

The Reasoning Trap

An ICLR 2026 paper proved that making AI agents reason harder actually makes them hallucinate more, not less. Toggle reasoning on and watch it happen.

Enhanced Reasoning

Toggle to see what happens when you turn on chain-of-thought / reasoning RL

Qwen 2.5-7B

Base Model · 7B params

Base
No Tool Available (should refuse)
34.8%
Distractor Tools (should pick correct tool)
54.7%

Qwen 3-8B

Base Model · 8B params

Base
No Tool Available (should refuse)
4.1%
Distractor Tools (should pick correct tool)
36.2%

Qwen 3-32B

Base Model · 32B params

Base
No Tool Available (should refuse)
5.1%
Distractor Tools (should pick correct tool)
46.6%

The Bottom Line

The layer in the neural network that should restrain bad tool calls is exactly what gets trained away during reasoning RL. The model learns to reason harder, but the guardrails against hallucination collapse in the process.

This isn't a bug in one model. It's a fundamental trade-off in how we train AI agents today. Every major reasoning method — RL, distillation, chain-of-thought — amplifies the same problem.

Enterprise Impact

47%

made decisions on hallucinated content

Agents in Production

96%

of enterprises run AI agents

Worst Case

2.1x

hallucination increase with reasoning

Source

"The Reasoning Trap: How Enhancing LLM Reasoning Amplifies Tool Hallucination"

ICLR 2026 · Rio de Janeiro · SimpleToolHalluBench

Read the full paper →

Building with AI agents?

We help teams ship reliable AI systems — with guardrails that actually work.

Talk to UsTry the Cost Calculator