TL;DR: Today's AI payment systems charge you per token like a parking meter ticking up costs. Users hate this. We're building plan-based pricing that works like Uber—you see a quote range, approve it, and pay for what's delivered. As AI costs drop over the next few years, this evolves into simple subscriptions. The future of AI payments is clearer pricing today, simpler pricing tomorrow.
The Problem: Micro-Transaction Friction
Imagine getting in an Uber and seeing the fare increase every few seconds as you drive. No upfront estimate. No idea when it'll stop. Just watching dollars tick up on your phone.
Most users would walk instead.
This is exactly how AI payment systems work today.
Every API call consumes tokens. Every token costs money. Users watch their credits drain as AI "thinks." They have no idea if the next request will cost them a dollar or fifty dollars. They hit arbitrary limits and get blocked mid-task. When they run out of credits, they're sent to a billing portal to upgrade—context completely broken.
The result? Most users either:
- Over-pay for subscriptions they don't fully use (concern about running out)
- Under-use AI tools (uncertainty about costs)
- Get blocked at critical moments (hit usage limits)
- Churn after hitting their first payment wall
Both sides are affected. Users feel nickel-and-dimed. AI companies lose customers at the exact moment they should be converting them.
How Real Services Actually Work
When you hire a contractor to remodel your kitchen, you don't pay them per hammer swing. Here's what actually happens:
Step 1: Discovery
- You describe what you want
- Contractor asks questions
- They inspect your space
- (This consultation is usually free or low-cost)
Step 2: Quote Range
- Contractor: "This will run you somewhere between $10K-$15K"
- They can't be exact (who knows what's behind those walls?)
- Range is honest—gives you upside and downside
Step 3: Approval
- You approve the maximum: "Okay, go ahead, but cap it at $15K"
- Now you have certainty
- Work can proceed without surprises
Step 4: Execution with Updates
- Contractor hits unexpected issue midway through
- "We found mold. Need another $3K to handle it properly."
- You decide: approve the extra cost or deliver what's done
Step 5: Final Delivery & Payment
- Work is complete
- Final cost: $13K (under the max!)
- You inspect, accept, and pay
- If quality is poor? You discuss before paying
This is how humans expect to pay for services. Not micro-charges per minute of labor.
What Users Actually Need
We talked to dozens of AI companies and their customers. Here's what users universally want:
1. Know Before You Go
"Just tell me what this is going to cost before I commit."
Users want a range. Not an exact number (impossible with AI), but a realistic estimate. Best case, expected case, worst case. Like Uber showing "This trip will be $15-$18" before you confirm.
2. Approve the Maximum
"Let me set a limit so I don't wake up to a surprise bill."
Users want to authorize a maximum spend. "I approve up to $20 for this analysis." The system can't exceed that without asking permission. Simple. Protective.
3. Get Notified When Approaching Limits
"Warn me before you hit my limit, don't just cut me off."
When the AI is approaching the approved maximum, notify the user:
- "Hey, I'm at 80% of your approved budget. Still good to continue?"
- Give the option: proceed, stop here, or approve more budget
No surprises. No sudden blocks.
4. Pay for Delivered Value, Not Failed Attempts
"I'm not paying for garbage output. I paid for results."
If the AI delivers low-quality work, users should have recourse. The contractor analogy holds: you don't pay full price for shoddy work. You rework or refund.
Plan-Based Pricing: How It Works
Here's our model. It mirrors how successful services actually price things:
Phase 1: Free Discovery
User describes their task:
"I need to analyze customer churn data and create a presentation."
AI asks clarifying questions:
- How many records?
- What time period?
- What insights matter most?
This conversation uses tokens, but it's covered by the user's subscription (or free tier). No charge for exploration. Like a contractor's free consultation.
Phase 2: Quote with Range
AI provides an honest estimate:
"Based on what you've shared, here's my plan:
Phase 1: Data Analysis
- Load and clean your dataset
- Run statistical analysis
- Identify churn patterns
Phase 2: Presentation Creation
- Generate slide outline
- Create visualizations
- Write presenter notes
Estimated cost: Somewhere between $8-$12
Why the range? Dataset cleanliness affects processing time. If your data is messy, I'll need to do more work.
Deliverables: Executive summary PDF, full analysis report, PowerPoint deck
Approve?"
No false precision. No hidden costs. Honest range based on complexity.
Phase 3: User Approval
User approves the maximum:
"Yes, proceed. My max is $12."
Now both sides have clarity. AI can execute without worrying about user surprise. User has certainty—won't exceed $12 without permission.
Phase 4: Execution with Checkpoints
AI works through the plan. If it's trending high:
"I'm about 70% done. Current cost is trending toward $11, within your approved range. Continuing..."
If something unexpected happens:
"I've hit a complexity issue. To deliver the full analysis, I need about $3 more (total $15). Want me to:
- Deliver what I have now for $10
- Approve the extra $3 for the complete version
Your call."
User stays in control. No surprise charges.
Phase 5: Delivery & Payment
AI completes the work. Final cost comes in under estimate:
"Analysis complete! Final cost: $9.50 (under your $12 budget).
Here are your deliverables:
- Executive summary
- Full analysis
- Presentation deck
Accept this work?"
If user accepts: charge $9.50 If user rejects (poor quality): discuss rework or partial refund
Pay for delivered value, not consumed compute.
Why This Model Works
For Users:
- No surprises: Know the max cost upfront
- Control: Approve budgets and overages explicitly
- Fair refunds: Don't pay for bad output
- Less anxiety: Can explore without micro-charging stress
For AI Companies:
- Higher conversion: Users approve plans instead of hitting hard blocks
- Better UX: Seamless upgrades at point of need
- Predictable revenue: Subscription base + usage overage
- Less support: Fewer "why was I charged?" tickets
For Developers:
- Standard patterns: One payment flow works across vendors
- Better DX: Clear API responses instead of custom error codes
- Easier integration: Consistent behavior across AI services
The Three Phases of AI Pricing Evolution
We're not building for today. We're building for where the market is going.
Phase 1: Today (Complex Pricing Required)
Reality:
- AI is expensive
- Models can't predict their own token usage well
- Quality varies
- Users need protection
Solution:
- Plan-based pricing with ranges
- Conservative estimates with safety nets
- Explicit approvals for large costs
- Refund policies for poor quality
This is necessary complexity because AI isn't commodity yet.
Phase 2: Near Future (1-2 Years)
What Changes:
- AI gets better at estimating its own work
- Costs drop significantly (50-75% lower)
- Quality becomes more consistent
- Models are faster
Result:
- Tighter estimation ranges
- Fewer approval interruptions
- More users fit within subscription tiers
- Simpler pricing becomes possible
The infrastructure we're building adapts automatically as AI improves.
Phase 3: End State (3-5 Years)
What Changes:
- AI is commoditized (like cloud compute today)
- Costs drop another 80-90%
- Estimation is highly accurate
- Quality is reliable
Result:
- Simple flat subscriptions work for 80% of users
- "Unlimited" plans become viable
- Complex pricing only for extreme edge cases
- AI becomes infrastructure, not a luxury
Think: AWS in 2010 (complex pricing, careful management) vs AWS in 2025 (most startups just use a simple tier).
Why Efficiency Always Matters
Even when AI is cheap, efficient usage matters. Here's our bet:
We'll always reward users who optimize:
You use AI efficiently? Get bonus credits. You share optimization techniques with the community? Earn rewards. You help us reduce waste? We pass the savings back to you.
Why? Because efficiency isn't just about cost:
- Speed: Fewer tokens = faster responses
- Quality: Concise prompts = clearer outputs
- Sustainability: Less compute = lower environmental impact
- Community: Sharing knowledge benefits everyone
As AI gets cheaper, the rewards shift from "save money" to "unlock features" and "community status." But the principle remains: we align with users who use AI thoughtfully.
Where Current Approaches Fall Short
Subscription-Only (Stripe Model)
How it works:
- Flat monthly fee
- Fixed limits (e.g., 100 API calls per month)
- Hard block when you hit limits
Problems:
- Over-pay if you don't use your allowance
- Under-serve power users (hit limits frequently)
- Awkward upgrade flow (go to billing portal mid-task)
Prepaid Credits (OpenAI Model)
How it works:
- Buy credits upfront
- Watch them drain per request
- Top up when you run low
Problems:
- Anxiety about running out
- Surprise depletion (didn't realize that request was expensive)
- Friction at top-up moment
Pay-Per-Token Micropayments (L402)
How it works:
- Charge exact token costs per request
- Lightning-fast tiny payments
- No subscriptions, pure usage
Problems:
- Death by a thousand cuts (every request is a charge)
- No cost predictability
- Terrible UX for humans (fine for agent-to-agent)
- Refunds are nearly impossible
Our Approach: Hybrid Plan-Based
How it works:
- Subscription base includes token allowance
- Plan quotes before execution
- Overage approved explicitly
- Pay for delivered value
Benefits:
- Predictable base cost (subscription)
- Flexibility for large projects (overage)
- User control (approval before big spend)
- Fair outcomes (refunds for poor quality)
What We're Building
Short-term:
- Plan-based pricing that works today
- Honest ranges instead of false precision
- User approval for budget overages
- Pay-for-value instead of pay-for-compute
Medium-term:
- Adaptive system that simplifies as AI improves
- Tighter estimates automatically
- Fewer interruptions as costs drop
- Community-driven optimization
Long-term:
- Simple subscriptions for most users
- Efficiency rewards regardless of cost
- AI becomes infrastructure
- Focus shifts to quality and speed
We're not just solving today's problem. We're building for the transition.
The Opportunity
Everyone else is building payment systems for expensive, unpredictable AI. They're optimizing for today's constraints.
We're building a system that adapts as AI evolves. Start complex (because reality is complex). Simplify automatically (as AI improves). Always reward efficiency (because thoughtful usage always matters).
This is a different bet.
Stripe built for recurring revenue and one-time charges. L402 built for micropayments. Traditional billing built for subscriptions.
We're building for the arc of AI evolution: expensive and unpredictable today, cheap and reliable tomorrow.
What This Means for You
If you're building an AI product, you need billing infrastructure that:
- Works today: Handles current high costs and estimation challenges
- Adapts tomorrow: Simplifies as AI gets better and cheaper
- Aligns incentives: Rewards users who help reduce costs
- Builds community: Shares optimization knowledge
If you're using AI products, you deserve pricing that:
- Respects your budget: Know costs upfront, approve maximums
- Values your time: No context-switching to billing portals
- Pays for results: Bad output shouldn't cost full price
- Gets simpler: As AI improves, pricing should too
This is where we're headed.
Join Us
We're building the first billing platform designed for AI's evolution from expensive to commodity. Plan-based pricing today. Simple subscriptions tomorrow. Efficiency rewards always.
Want to be part of this?
- Join our early access program - Be among the first to use plan-based AI pricing
- Read our technical deep-dive - How HTTP 402 and L402 fit into this vision
- See why current UX fails - The user experience problems we're solving