See the signals shaping your next planning cycle.

The plan is due and the world keeps changing. Signals tracks the forces that could reshape the next 12 to 36 months and flags the ones your team should discuss. You get a radar with sources on every signal.

Request scan

What is a horizon scan?

A horizon scan is a structured look at policy, science, industry, and media for early signals of change. It looks 12 to 36 months ahead, so your team can spot a shift while there is still time to respond.

What the scan does, and what your team keeps.

What it does

Finds signals in literature, policy filings, scientific publications, and credible media.
Groups them by category and scores verifiability. Signals with support from several sources rise to the top.
Refreshes on a schedule so the radar keeps pace with the world.

What you keep

A categorized radar of 40 to 100 signals your team can read in five minutes.
Sources on every signal, so each claim has evidence behind it.
A Workspace you can revisit and rerun each quarter.

How it groups what it finds.

The categories this scan sorts signals into by default. Rename, merge, or replace them once the Workspace is yours.

SocialTechnologicalEconomicEnvironmentalPolitical

See an example from this scan.

Run on the theme AI infrastructure scaling across 33 models. 213 unique signals after dedup, 0 confirmed by two or more models.

Open the full example

AI infrastructure scaling

40 signals · best of 33

ComputeModelsToolingEconomics

The signals it surfaces.

These signals came from runs of this scan. Each one is merged across models, scored, and checked against sources, so your team can inspect the evidence behind it.

Research prompt

What is changing across Compute scaling limits, inference economics, and the post-training tooling stack?

SignalGrounded

Verifiability

Specificity

Inference cost per token decline

OpenAI and Anthropic reduce inference costs by 50% in 2023. Signals a competitive pricing war for AI services.

Merged from 2 model signals

“Inference Cost Per Token Decline”
Claude Opus-4.6
“Inference cost per token decline”
GLM 4.6

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Data center power grid constraints

Utility providers deny power allocation requests for new AI training clusters. Indicates geographic compute distribution depends on energy availability rather than latency.

Merged from 2 model signals

“Data center power grid constraints”
GLM 5.1
“Data center power grid constraints”
Sonar Reasoning-Pro

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Observability for LLM pipelines

Arize and Weights & Biases add prompt drift detection. Signals need for real-time monitoring in production deployments.

Merged from 2 model signals

“Observability for LLM pipelines”
Qwen Max
“Observability for LLM pipelines”
Mistral Large-2512

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Observability for LLM pipelines

Arize and Weights & Biases add prompt drift detection. Signals need for real-time monitoring in production deployments.

Merged from 2 model signals

“Observability for LLM pipelines”
Qwen Max
“Observability for LLM pipelines”
Mistral Large-2512

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Inference cost per token decline

OpenAI and Anthropic reduce inference costs by 50% in 2023. Signals a competitive pricing war for AI services.

Merged from 2 model signals

“Inference Cost Per Token Decline”
Claude Opus-4.6
“Inference cost per token decline”
GLM 4.6

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Data center power grid constraints

Utility providers deny power allocation requests for new AI training clusters. Indicates geographic compute distribution depends on energy availability rather than latency.

Merged from 2 model signals

“Data center power grid constraints”
GLM 5.1
“Data center power grid constraints”
Sonar Reasoning-Pro

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

SignalGrounded

Verifiability

Specificity

Observability for LLM pipelines

Arize and Weights & Biases add prompt drift detection. Signals need for real-time monitoring in production deployments.

Merged from 2 model signals

“Observability for LLM pipelines”
Qwen Max
“Observability for LLM pipelines”
Mistral Large-2512

Converged across 2 of 33 frontier models on “AI infrastructure scaling.”

When to use this scan.

Strategy leads
Building a strategic plan and want to test the assumptions against external signals first.
Strategy leads, planning teams
Founders & chiefs
Briefing a board on what's changing that the team isn't paying attention to yet.
Chiefs of staff, CSOs
Foresight & R&D
Setting up an ongoing intelligence function rather than commissioning one-off reports.
Foresight units, R&D

Free scan

Try a horizon scan on your own theme.

Your first scan is on us.