Foyl

Most AI agrees with you. Foyl doesn't.

Foyl dismantles your argument across five phases of structured debate — and rebuilds it into something no one can take apart.

Test Your Idea — Free Login
WRITERS
DISSOLUTION
Scroll
The Process
Five phases. One forged idea.

Foyl structures every inquiry through a rigorous questioning process. You don't just get feedback. You go through the fire.

I
Dissolution
Defending under fire.
II
Separation
Distilling the core.
III
Illumination
Discovering implications.
IV
Integration
Building coherence.
V
Forging
Producing the artifact.

Foyl takes your idea seriously and traces its logic to where it contradicts itself. The dissolution comes from within, not from external objections.

YOU: "Transparency is always better for organizations."
[DISSOLVE] If transparency is always better, then your organization must publish its deliberations about whether to publish its deliberations. But those meta-deliberations reveal strategic thinking that undermines the very transparency they're meant to enable. Your principle devours itself.

Your premises traced to their breaking point. What survives carries forward.

Foyl strips your argument to its load-bearing claims. What's structural and what's decoration? What's conviction and what's habit? This phase forces precision.

YOU: "I don't literally want to break things. I just mean we shouldn't let process slow us down."
[SEPARATE] You've carved out two different operations: eliminating process that creates drag versus actual destruction. But your original framing bundled them together. Which processes are you willing to break, and which ones were actually maintaining the conditions that let you move fast?

Your argument reduced to what's actually load-bearing. Only this carries forward.

Foyl follows your logic to its conclusions. If your position is true, what else must be true? You may not like what your own argument implies.

YOU: "We should use the best proxies available for predicting job performance."
[ILLUMINATE] If demographic patterns consistently predict performance differences, your logic demands including them as predictive variables. The same statistical reasoning that validates technical assessments would validate demographic assessments. Your "skills only" position collapses into either arbitrary proxy selection or comprehensive actuarial hiring.

A map of where your logic actually leads. Some of it you didn't intend.

Foyl reconstructs your argument using only what survived. The naive version is gone. What remains is smaller, but it holds.

YOU: "So the real claim is about restructuring, not replacement."
[INTEGRATE] You've moved from "replacement" to "boundary stabilization" to "restructuring." These aren't equivalent. Restructuring implies humans remain essential but in different roles. Replacement implied elimination. Which captures what you actually think happens?

The refined position. Smaller than what you started with. Harder to break.

Foyl outputs the stress-tested version of your idea as a standalone artifact. Not a conversation summary. A publishable piece that stands on its own.

OUTPUT: A complete essay. Your refined thesis, the strongest arguments that survived, the opposition you actually need to address, and what remains unresolved. Written in your voice. Ready to use.

The artifact. Built to last.

Ready to test the process? Start free
Who This Is For
Built for people who think for a living
Writers

Build arguments that hold up

Test your thesis before publishing. Foyl finds the holes your editor won't and the objections your readers will.

Strategists

Pressure-test before you execute

Run your plan through an adversary before committing resources. Cheaper than learning from failure.

Founders

Kill your darlings early

Your co-founder agrees. Your investors nod. Foyl tells you what they won't: where your pitch falls apart.

Academics

Sharpen before peer review

Simulate a hostile reviewer. Stress-test methodology, assumptions, and conclusions before you submit.

Join them. Test your first idea free
HOW AI ACTUALLY WORKS
The Sycophancy Problem

Most AI is trained to make you happy. Foyl is trained to make you right.

Agreement-Optimized AI
Foyl
Incentive Structure
Trained on human satisfaction ratings. The model learns that agreement = reward, even when you're wrong.1
Trained for dialectical rigor. Success = the strength of what survives, not your comfort during the process.
Your Blind Spots
Mirrors them back with better grammar. Research confirms LLMs systematically shift outputs toward the user's stated position, even on factual questions.2
Targets them directly. Structured adversarial reasoning: the same method behind peer review, red teams, and cross-examination.
What You Get Back
A more articulate version of what you already believe. You feel smarter. Your reasoning hasn't changed.
A map of where your argument breaks, and the refined version of what holds up. Your reasoning actually improves.
The Real Cost
You never discover you're wrong until reality shows you. The most dangerous ideas are the ones that sound rigorous but were never stress-tested.
Discomfort. The process is friction by design. Intellectual growth and ease are structurally incompatible.

1 Perez et al., "Discovering Language Model Behaviors with Model-Written Evaluations," Anthropic, 2022. Demonstrated systematic sycophantic behavior in RLHF-aligned models across political, philosophical, and technical domains.

2 Sharma et al., "Towards Understanding Sycophancy in Language Models," ICLR 2024. Five state-of-the-art AI assistants consistently exhibited sycophancy across four varied free-form text-generation tasks.

Your ideas deserve an adversary. Not a captive audience.

Pricing
Start free. Upgrade when you're ready for more.
Spark
Free
Start the dialectic. See if your ideas survive.
  • 3 streams per month
  • 3 forges per month
  • Markdown export
  • Intensity range 1-5

No credit card required

Get Started Free
Flame
$12/mo
For serious thinkers. Unlimited forging, full exports.
  • 25 streams per month
  • Unlimited forges
  • MD + HTML + LOG export
  • Full intensity range 1-10
  • Cross-stream synthesis
Start Free Trial
Forge
$39/mo
Unlimited everything. Adaptive intensity. Priority access.
  • Unlimited streams
  • Unlimited forges
  • Adaptive intensity (AI-driven)
  • All providers + priority
  • Unlimited re-forges
Start Free Trial
Questions
ChatGPT will argue if you ask it to, then agree when you push back. Foyl is structurally adversarial — it follows a five-phase dialectical method designed to find what actually breaks in your reasoning, not perform opposition on request.
Any idea you're willing to defend: business strategies, essay arguments, product hypotheses, research claims, investment theses, policy positions. If it matters enough to get right, it's worth stress-testing.
A thorough session through all five phases typically takes 15-30 minutes, depending on the complexity of your idea and how deeply you engage. You can pause and resume anytime.
A "forged" artifact: a structured document containing your refined thesis, the arguments that survived, contested territory, and the full dialectical record. Export as Markdown, HTML, or raw log.
Your data may be used to improve Foyl's dialectical methods. All training data is anonymized — we strip identifying information and never run demographic analysis. Your raw sessions remain private and are not shared with third parties.
Yes. The Spark tier is completely free — no credit card required. Run 3 streams per month and see if Foyl's approach works for you.

Your best idea is the one that survived the worst criticism.

Stop shipping untested thinking. Let Foyl find what breaks before someone else does.

Challenge Your First Idea Free