# AI That Thinks Through Problems

> Chain-of-thought reasoning at the edge with <200ms latency. Deploy AI agents that break down complex problems, explain their reasoning, and make intelligent decisions with transparent logic chains.

Canonical: https://thinking.do/

## In code

```ts
import { Thinking } from 'thinking.do'

// Chain-of-thought reasoning at the edge
const analyst = Thinking.agent({ role: 'analyst', model: 'claude-sonnet-4' })

on.Revenue.threshold.exceeded($ => {
  const decision = Thinking.reason({
    observation: $.data,
    options: ['scale_marketing', 'hire_sales', 'expand_regions'],
    criteria: ['ROI', 'risk', 'impact']
  })
  $.Strategy.execute(decision)
})

Thinking`reduce churn by 30% within budget of $100K`
```

## Frequently asked questions

### Why use thinking.do instead of calling AI models directly?

Thinking.do specializes in reasoning tasks with optimized prompts for chain-of-thought processing, automatic reasoning validation, and transparent decision explanations. Direct API calls lack structured reasoning frameworks, validation steps, and explainability features. Edge deployment provides <200ms latency globally with automatic caching of similar reasoning patterns (80% cost reduction). Multi-model ensemble selects best reasoning model for each problem type (analytical, creative, technical).

### How do you ensure reasoning quality and accuracy?

Multi-step validation with consistency checking, fact verification against knowledge bases, and confidence scoring for each reasoning step. Low-confidence conclusions trigger additional analysis or human review. Reasoning chains are auditable with full provenance tracking—see exactly how agent reached each conclusion. A/B testing different reasoning strategies measures real-world impact. Human feedback loop improves reasoning quality over time with reinforcement learning.

### What types of problems can thinking.do solve?

Complex analysis (market research, competitive analysis), strategic planning (goal decomposition, resource allocation), decision-making (multi-criteria evaluation, trade-off analysis), problem diagnosis (root cause analysis, troubleshooting), creative ideation (brainstorming, solution design), and logical reasoning (mathematical proofs, code debugging). Handles problems requiring 5-10 reasoning steps with maintained coherence. Specialized modes for different reasoning types: analytical, creative, strategic, technical.

### Can reasoning agents collaborate and learn?

Yes. Multiple reasoning agents can collaborate via semantic protocols to solve complex problems requiring diverse expertise. Agents debate solutions, challenge assumptions, and synthesize conclusions through structured dialogue. Reasoning patterns are learned and improved over time—agents remember successful approaches and avoid failed strategies. Transfer learning applies reasoning from one domain to similar problems. Collective intelligence emerges as agent network grows.

### How transparent and explainable is the reasoning?

Full transparency with human-readable reasoning chains showing each logical step, assumptions made, evidence considered, and alternatives rejected. Confidence scores for each step help identify weak reasoning. Visual reasoning graphs show relationships between concepts and decisions. Natural language explanations tailored to audience (technical, executive, non-technical). Counterfactual analysis shows how different inputs would change conclusions. Export reasoning for compliance and audit requirements.

## Part of Platform.do

thinking.do is one of the primitives of Platform.do, in the Agents layer (reason, research and act). Every .do shares one grammar, Business-as-Code, and one SDK.

- What is Business-as-Code? https://platform.do/business-as-code
- Platform.do: https://platform.do/
- Docs: https://docs.platform.do