An always-on intelligence foundation
Scheduled production collection brings selected signals together. Max maintains evidence-linked findings and durable cases, identifies investigation leads, and consults its operational specialist layer.


Meet Max. Here to connect the dots, help the team, and find the next useful step. Today, he brings signals and investigations together. The ambition: coordinate every agent and help BestPrice keep getting better.
Specialists & bounded checks
Findings & outcome memory
A stream of signals. Thousands of possible next moves. One intelligence learning which small decision matters now.
The ambition for Max is to become the coordinating intelligence across BestPrice: a persistent understanding of what is happening, what matters, who can help, and whether an action actually worked.
The story unfolds through micro-decisions: inspect this signal, defer that duplicate, ask a specialist, gather one missing fact, or wait for better evidence. Over time, those small choices could coordinate operations, engineering, product and customer experience around the same evolving picture.
This is the long-term vision. The production foundation below is the starting point.
Max runs in the production backend. Its current job is to observe, investigate and remember. Broad control over BestPrice is the long-term ambition.
Scheduled production collection brings selected signals together. Max maintains evidence-linked findings and durable cases, identifies investigation leads, and consults its operational specialist layer.
Decide, delegate, verify, take a bounded memory action and learn from the result—including reuse and withdrawal.
Own a real investigation through a measured outcome, then earn permission for narrowly scoped, reversible repairs.
Max can generate investigation leads and proposed plans. It does not yet autonomously repair BestPrice or demonstrate verified business impact. Model-proposed actions remain disabled.
From code and errors to customer signals and external incidents, Max is building a common evidence layer across tools that usually tell only part of the story.
Track registered work, changes and reported failures. A heartbeat alone is not proof of health.
Connect selected code context, repository activity and error metadata to investigation leads.
Bring sampled work and support signals into the same evidence picture.
Inspect bounded metrics and measurement coverage before interpreting changes.
“See all” is the direction. Today, Max sees selected surfaces—not every event, every agent or every process. Optional connections are only active when configured.
Internal telemetry tells part of the story. External incidents, changing dependencies and new information add context. The ambition is a continuous cycle of attention, delegation and learning.
A provider publishes an incident. An error pattern emerges. A task changes. A measurement shifts.
Today: selected internal signals and public provider feeds are collected.Is it fresh? Relevant to a known dependency? Already being investigated? Is another check worth its cost?
Today: fixed checks use freshness, backoff and budget rules; leads retain uncertainty.“Inspect this dependency. Return the evidence, what is missing, and the condition that would change our conclusion.”
Today: bounded specialist checks. Future: coordinated missions across specialist agents.Verify the answer, choose the smallest justified next step, and record whether it helped.
Proven: bounded memory actions. Future: authorized operational actions and useful notifications.A provider announces disruption. Max checks whether a known dependency is relevant, asks an operations specialist to inspect the affected path, and requests independent evidence of impact. If nothing supports a connection, it waits. If impact is verified, it chooses an authorized response and measures recovery.
The value is in choosing the next useful step—not turning every signal into an alarm.What should I inspect? Who should investigate? Is the answer trustworthy? Should I act, wait or stop? Max’s future is built from these micro-decisions. Explore the narrow loop already demonstrated below.
Start from an unresolved case. Weigh evidence, freshness and uncertainty, then select an allowed investigation within a fixed budget.
Illustrative walkthrough of an executed proof of concept. These controls do not run production actions. “Learn” means recorded evidence and outcomes—not model retraining.
Max is your BestPrice AI teammate: a friendly conversation with a growing understanding of the business. Ask a question, share a problem, or find out what needs attention.
Ask about operations, products, measurements and the work behind them. Max brings the existing assistant’s capabilities into the intelligence initiative.
max status — the latest stored picture.max cases — open investigations.max decisions — recorded choices.max help — where to begin.
Try /max-help to explore the tools, /max-status for connection status, /max-deal for current deals, and /max-nameday for today’s celebrations. Use these in Slack; this page is the public overview.
Useful intelligence narrows uncertainty. It also knows when the evidence is not enough.
Before calling a change in search performance good or bad, Max investigates whether eligible impressions and redirects can actually be linked.
Inspect aggregate measurement coverage and ID presence.
Separate traffic groups. A page label alone does not establish the instrumented search path.
Verify the served route and follow the same real ID through its own journey. Recovery remains unproven.
Max is intended to know which agents exist, what they can do, what they are working on, and what evidence they need next. It would coordinate the mission from beginning to end.
Inspect services and dependencies
Investigate code and prepare repairs
Reproduce faults and verify outcomes
Test hypotheses against measurements
Connect friction to product behavior
Investigate opportunities and external change
One case. Clear ownership. Linked evidence. Action receipts. Lessons the next agent can use.
Direct read-only specialist interaction and a bounded learning loop are demonstrated. General agent spawning, company-wide coordination and autonomous repairs are planned capabilities. These specialist roles illustrate that destination.
Six concrete ways the foundation could grow into an operating capability. Each example separates today’s tools from the future action—and defines what success would need to prove.
A new error pattern appears alongside a release and a change in service performance.
Selected Sentry, repository and performance signals can support an investigation lead. A shared timeline is evidence to investigate, not proof of cause.
Coordinate a code investigator and a test agent, reproduce the fault, prepare a fix, and apply it only within an approved scope.
Verify the affected behavior and error rate over a declared window. Roll back if the outcome is worse.
A registered job stops showing useful progress or reports repeated failures.
Inspect process signals and request a bounded, read-only check. Keep the unresolved case and the reply.
Delegate diagnosis, identify dependencies, and execute an allowed retry or recovery runbook when its conditions are met.
Confirm useful output resumed—not just that a process restarted.
A measurement suggests fewer shoppers are progressing from search to merchant offers.
Investigate measurement coverage, attribution IDs and traffic populations before interpreting the change.
Coordinate analytics and product agents to distinguish an instrumentation gap from a relevance or experience problem, then test a targeted improvement.
Compare eligible, matched cohorts and account for reporting delay. Demonstrate a real effect.
Support activity and application errors may point to the same broken experience.
Collect selected support metadata and error signals. Surface leads that need stronger evidence.
Bring a support investigator and an engineering agent into one case, establish the connection, prepare a repair and draft a useful update.
Verify the customer journey works again. Send updates only through authorized channels.
A provider reports an incident while dependent operations show unusual behavior.
Read selected public provider feeds and explicit dependency relationships, alongside internal signals.
Estimate the affected scope, coordinate specialist checks, and choose an approved fallback or escalation when the evidence supports it.
Verify the affected path recovers. Do not mistake a provider status change for internal recovery.
Several agents may be investigating the same issue, waiting on each other, or repeating expensive checks.
Use durable cases, check budgets and reusable verification records. Current agent visibility is partial.
Maintain a shared agent registry, assign missions, manage dependencies and resource budgets, stop duplicate work, and reuse verified lessons across teams.
Measure completion quality, investigation cost and time to verified resolution—not agent activity alone.
Illustrative use cases, not reports of completed incidents or promised business results.
A company that notices what needs attention, brings the right intelligence to the problem, acts within its mandate, and learns from the result. This is the destination for Max. Reliability, measured value and human accountability determine how far it can go.
Connect selected signals. Preserve cases. Consult specialists. Generate evidence-backed leads.
Proof: scheduled production runs and bounded investigations.Carry an investigation from question to verification, with quality, effort and coverage recorded.
Gate: demonstrate a real, measured resolution.Spawn or select specialist agents, coordinate their work, apply approved reversible actions and verify their effects.
Gate: scoped permissions, rollback and independent outcome checks.Anticipate emerging problems, discover opportunities and continuously coordinate improvements across BestPrice, guided by shared evidence and measurable goals.
Gate: sustained reliability, measured value and human accountability.
Max starts with awareness. Its ambition reaches across every agent, every process and every decision that can make BestPrice better. The foundation is running. The mission is much bigger.