M4 · PROTOTYPE TO PRODUCTION · P12

The Anthropic Agent Environment Playbook

Prototype in Perplexity. Run it here when it has to run by itself.

10 SLIDES · ~10 MIN · PLAYBOOK OVERVIEW
Key Takeaways
  • Prototype in Perplexity. Run in the Agent Environment.
  • $0.08 — per session-hour, plus token rates. Managed sandbox pricing.
  • 10× — faster agent delivery cited by Anthropic for early launch partners.
  • 30 d — from kickoff to first production deploy on a managed runtime — our recommended target.
  • Identify the one stalled agent that should be running autonomously but is not — that is your migration candidate.
MODULE 4 · PLAYBOOK 12

The Anthropic Agent Environment

Prototype in Perplexity. Run it here when it has to run by itself.

BuildClub Academy
01 · THE HOOK

The infrastructure barrier to running real agents just dropped.

$0.08
per session-hour, plus token rates. Managed sandbox pricing.
Anthropic Managed Agents
10×
faster agent delivery cited by Anthropic for early launch partners.
Anthropic launch announcement (Apr 2026)
30 d
from kickoff to first production deploy on a managed runtime — our recommended target.
BuildClub field interviews · 2026
02 · THE THESIS

Prototype in Perplexity.
Run in the Agent Environment.

The Thesis
03 · WHY MIGRATE

Six things you get here that you cannot get in a chat session.

1
2
3
4
5
6
1
Secure sandboxing
Agent runs in its own isolated runtime, not in your browser tab.
2
Long-running sessions
Work survives past chat turns. Agents can think and act for hours.
3
Multi-agent coordination
A coordinator dispatches work to worker agents and synthesizes results.
4
Agent Vault
Dedicated secret storage per agent. Credentials scoped and rotated automatically.
5
Memory Store
Persistent context across sessions. The agent remembers what it learned last week.
6
Outcomes Grader
Built-in eval framework. Define success criteria, the system grades outputs automatically.
04 · THE MIGRATION ARC

What moves from prototype to production — named, mapped, decoupled.

STARTING POINT PROTOTYPE prompts · tools · loops END STATE PRODUCTION governed · observable · safe prompt → Skill tools → Toolset memory → Store loop → Harness Same agent. New scaffolding. Four named upgrades.

Prototype → Production is a promotion, not a rewrite. Skills, toolsets, vault, memory, eval — each prototype concept has a named home on the managed runtime.

05 · ANTHROPIC’S SIX TERMS

Six terms. Use Anthropic’s vocabulary internally — then map your vendor.

Skills
The named, versioned capability an agent can call. The atomic unit of agent behavior.
Toolsets
The grouped collection of tools available to a skill. Scoped per agent.
Agent Vault
Per-agent secret store. Credentials never shared across agents.
Memory Store
Persistent context that survives sessions. The "what it learned" layer.
Outcomes Grader
The eval framework. Defines success criteria and grades outputs automatically.
Coordinator
The orchestrator that dispatches work to sub-agents and synthesizes results.
06 · PRICING

What it costs — and when to migrate vs stay in Perplexity.

Takeaways
  • 1 $0.08 per session-hour (managed sandbox compute) plus standard Anthropic token rates.
  • 2 Migrate when: agent runs autonomously / serves multiple users / needs long-running sessions / needs persistent memory.
  • 3 Stay in Perplexity when: you are the only operator / prompt-driven workflow / prototype phase / thesis not yet proven.
07 · THE MIGRATION CHECKLIST

Eight items before your first production deployment.

1
2
3
4
5
6
7
8
1
Thesis proven
Prototype has measured metric movement.
2
Job spec written
Six-field spec from Agent Migration Playbook complete.
3
Skills defined
Every prompt is named and versioned.
4
Toolsets scoped
Minimum tool access documented and approved.
5
Agent Vault provisioned
No shared service accounts. Per-agent credentials only.
6
Memory Store seeded
Initial context loaded. Refresh cadence defined.
7
Outcomes Grader configured
Success criteria defined. Baseline established.
8
Blast radius limit set
What the agent cannot do without human approval.
08 · EARLY ADOPTER RESULTS

Two launch partners. One pattern.

1
Rakuten — vendor-reported
Anthropic's launch documentation reports Rakuten cut critical errors by 97% and moved releases from quarterly to biweekly on the managed runtime. Treat as a vendor-reported figure — verify the math against your own pilot before quoting it to your board.
2
Notion — named launch partner
Migrated agents that draft and route project work onto the managed runtime in the first weeks after launch. Anthropic also names Asana, Atlassian, and Sentry as launch partners on the same pattern.
MONDAY-MORNING ACTIONS

Four moves to run your first production agent in 30 days.

Monday Morning
  1. 1
    Identify the one stalled agent that should be running autonomously but is not — that is your migration candidate.
  2. 2
    Map its skills and toolsets against the six-service vocabulary — name each one.
  3. 3
    Provision the Agent Vault — create per-agent credentials, revoke any shared service accounts.
  4. 4
    Ship to the Agent Environment in the next sprint — with Outcomes Grader configured before the first production run.
P12 · Questions CEOs Ask

Frequently Asked Questions

What is the core idea of The Anthropic Agent Environment Playbook?
Prototype in Perplexity. Run in the Agent Environment.
What does the data say a CEO should pay attention to?
$0.08 per session-hour, plus token rates. Managed sandbox pricing. 10× faster agent delivery cited by Anthropic for early launch partners.
What should a CEO do Monday morning after reading The Anthropic Agent Environment Playbook?
Start here: Identify the one stalled agent that should be running autonomously but is not — that is your migration candidate; Map its skills and toolsets against the six-service vocabulary — name each one; Provision the Agent Vault — create per-agent credentials, revoke any shared service accounts.
What are the steps in The Anthropic Agent Environment Playbook?
1) Secure sandboxing; 2) Long-running sessions; 3) Multi-agent coordination; 4) Agent Vault; 5) Memory Store.
Where do the claims in The Anthropic Agent Environment Playbook come from?
The playbook cites Anthropic Managed Agents announcement (Apr 2026); Anthropic Managed Agents pricing & launch (Apr 8, 2026); Anthropic Managed Agents documentation (Apr 2026); Anthropic Managed Agents documentation.
P12 · Sources

All Sources in This Playbook

00 / 10