Structured meta-thinking assistant
See how the assistant frames the task before you trust the answer.
SecondOrder turns hidden orchestration into a product surface. In `/chat`, complex requests can show task framing, a compact plan preview, confidence signals, and structured feedback so the assistant is easier to inspect and improve.
Meta mode
Visible framing
Workflow
Planner + critic
Evaluation
Feedback + events
Capabilities
SecondOrder v1Visible meta mode
Meta-routed replies show the detected task type, a compact goal summary, constraints, and whether SecondOrder is using a structured pass.
Plan preview
Complex requests include a compact plan preview that stays collapsed by default and expands when you want to inspect the approach.
Confidence signals
Each meta response can surface confidence, limitations, and context gaps so the assistant is easier to trust and easier to challenge.
Feedback and evaluation
Assistant messages support structured feedback, and the chat lifecycle emits evaluation events for product-level measurement.
What is it
A deliberate, adaptive layer for reasoning.
SecondOrder is a chat product that exposes enough of its meta layer to make harder requests legible. It classifies the task, builds structured workflow output, and surfaces the parts users actually need to evaluate the answer.
A meta layer on top of chat that makes task interpretation visible.
Structured planner and critic passes running behind the assistant.
Compact plan previews for planning, analysis, decision, and troubleshooting tasks.
Thread-scoped history and resource isolation across conversations.
Event instrumentation for completion, feedback, and plan-preview usage.
Implemented today
A working v1 is already in place.
SecondOrder is no longer just positioning. The current release already ships the first user-visible version of the meta-thinking experience inside chat.
Meta UX
Task framing, plan previews, and trust signals
The shipped chat experience shows how SecondOrder framed a meta-routed task before or alongside the final answer.
Workflow
Planner and critic passes in production
Complex requests already run through structured planning and critique steps that produce goal, constraints, plan, confidence, and limitation metadata.
Learning loop
Structured feedback and evaluation events
Users can rate individual assistant messages, and the product records core lifecycle events for measuring outcomes over time.
Chat foundation
Threaded history and onboarding
The current release preserves thread URLs, loads chat history per thread, and explains the kinds of problems SecondOrder handles well before the first message.
Meta-thinking path
A visible diagram of how SecondOrder thinks.
The system does not run the full planner and critic stack for every prompt. It first classifies the request, then chooses the light path or the structured path based on how much reasoning the task needs.
Why this matters
Friends trying the product can understand it as a chat system with a decision gate: answer directly when the task is easy, or switch into a planning and critique loop when the task is ambiguous, risky, or multi-step.
The key product idea is the decision gate in the middle: simple prompts stay fast, while complex prompts trigger the planning and critique loop.
Iterative problem solving
A structured loop designed for harder requests.
Interpret the task and decide whether meta mode is needed.
Generate a compact plan and response strategy for harder requests.
Review the draft for confidence, limitations, and missing context.
Return the answer with visible framing, then capture user feedback.
Self-auditing
Trust signals that expose uncertainty without dumping internals.
SecondOrder keeps the internal prompts hidden, but it can surface confidence, limitations, and context gaps so the answer is easier to inspect before you act on it.
Workflow
Planner + critic
User signal
Feedback captured
Meta-thinking in AI
Cognition that can inspect itself.
Inspired by human metacognition, SecondOrder makes planning, critique, and uncertainty visible enough to build trust without exposing raw chain-of-thought.
Task classification routes complex requests into a structured meta pass.
Planner and critic workflow output drives the UI instead of freeform guesswork.
Trust signals stay compact so the chat still feels fast and conversational.
Product roadmap
What comes after structured meta chat.
The current release productizes planning and critique inside chat. The next releases push deeper into tools, evidence-aware execution, memory, and orchestration.
Shipped now
Structured meta chat
Visible task framing, planner and critic workflow output, compact plan previews, confidence signals, structured feedback, and lifecycle instrumentation are live.
Next
Tools and evidence-aware execution
The next step is moving from reflective chat into reflective execution with tool routing, evidence-aware answering, and stronger stop conditions.
Later
Memory and orchestration
Longer term, SecondOrder can add durable memory, adaptive strategy reuse, and a coordinator layer for multi-pass or multi-model execution when justified.