Changelog

A log of every change to this wiki, newest first, grouped by month. Each page shows up here when it is added (New), every time it is edited (Updated), and when it is deleted (Removed). Each row shows the date, the section, the title, and the one-line description from frontmatter.

Dates and changes are derived from git history (renames followed). Changes that have not been committed yet will not appear here.

This is the full version of the "recent changes" list on the home page.

July 2026

Updated 2026-07-22 (Concept) Consumption Pricing: The Claude Developer Platform bills by tokens consumed, not seats sold. A builder's bill scales with how much their product actually runs, which aligns the platform with usage that works.
Updated 2026-07-22 (Concept) Memory Stores: Workspace-scoped persistent memory for Managed Agents: a filesystem mounted into every session, versioned, with an audit trail, so what an agent learns survives the run.
Updated 2026-07-22 (Concept) Outcomes: The Managed Agents primitive for defining done: you supply a rubric, and the platform runs an iterate, grade, revise loop with a separate grader until the work passes.
Updated 2026-07-22 (Concept) Scheduled Deployments: Cron for agents on Managed Agents: a schedule fires a fresh session on the clock, every firing leaves a run record, and the whole thing can be paused or archived.
Updated 2026-07-22 (Concept) Self-Correction Mid-Task: An agent that checks its own output against reality before handing it over, and fixes what it finds: validate in a sandbox, look at the rendered result, retry on failure.
Updated 2026-07-22 (Concept) The Audit Trail: Citation to source as a product feature: every answer carries a pointer to the exact place it came from, so the human verifies instead of hunts.
Updated 2026-07-22 (Concept) The Batch API: 50 percent off input and output tokens for anything that can wait up to 24 hours. The discount stacks with prompt caching.
Updated 2026-07-22 (Concept) The Effort Dial and Model Splitting: Spend frontier-model tokens on judgment and small-model tokens on commodity steps, and turn the effort parameter down where deep thinking is wasted.
Updated 2026-07-22 (Concept) The No-API Integration: Using computer use to operate the screens of systems that will never get an API. The largest architecture requirement in the newest YC batch, and a proven pattern in the customer corpus.
Updated 2026-07-22 (Concept) Vaults: The Managed Agents credential store: secrets never enter the sandbox, they are substituted into outbound requests at egress, so a prompt injection has nothing to steal.
Updated 2026-07-22 (Concept) Consumption Pricing: The Claude Developer Platform bills by tokens consumed, not seats sold. A builder's bill scales with how much their product actually runs, which aligns the platform with usage that works.
Updated 2026-07-22 (Concept) Memory Stores: Workspace-scoped persistent memory for Managed Agents: a filesystem mounted into every session, versioned, with an audit trail, so what an agent learns survives the run.
Updated 2026-07-22 (Concept) Outcomes: The Managed Agents primitive for defining done: you supply a rubric, and the platform runs an iterate, grade, revise loop with a separate grader until the work passes.
Updated 2026-07-22 (Concept) Scheduled Deployments: Cron for agents on Managed Agents: a schedule fires a fresh session on the clock, every firing leaves a run record, and the whole thing can be paused or archived.
Updated 2026-07-22 (Concept) Self-Correction Mid-Task: An agent that checks its own output against reality before handing it over, and fixes what it finds: validate in a sandbox, look at the rendered result, retry on failure.
Updated 2026-07-22 (Concept) The Audit Trail: Citation to source as a product feature: every answer carries a pointer to the exact place it came from, so the human verifies instead of hunts.
Updated 2026-07-22 (Concept) The Batch API: 50 percent off input and output tokens for anything that can wait up to 24 hours. The discount stacks with prompt caching.
Updated 2026-07-22 (Concept) The Effort Dial and Model Splitting: Spend frontier-model tokens on judgment and small-model tokens on commodity steps, and turn the effort parameter down where deep thinking is wasted.
Updated 2026-07-22 (Concept) The No-API Integration: Using computer use to operate the screens of systems that will never get an API. The largest architecture requirement in the newest YC batch, and a proven pattern in the customer corpus.
Updated 2026-07-22 (Concept) Vaults: The Managed Agents credential store: secrets never enter the sandbox, they are substituted into outbound requests at egress, so a prompt injection has nothing to steal.
New 2026-07-22 (Concept) Long-Running Work: Agent runs measured in hours, not seconds: the workload class that makes the harness and deployment decisions matter, with published runs of four, six, and seven hours.
New 2026-07-22 (Concept) Self-Correction Mid-Task: An agent that checks its own output against reality before handing it over, and fixes what it finds: validate in a sandbox, look at the rendered result, retry on failure.
New 2026-07-22 (Concept) Silent Failure: The failure mode that kills trust in unattended agents: the agent thinks it finished, and it did not. Loud failure is recoverable; silent failure is the trust problem.
New 2026-07-22 (Concept) The Audit Trail: Citation to source as a product feature: every answer carries a pointer to the exact place it came from, so the human verifies instead of hunts.
New 2026-07-22 (Concept) The Company Brain: A persistent store of how a specific company works that agents read on every run and write back to as they learn. The pattern an entire YC batch was asked to build.
New 2026-07-22 (Concept) The No-API Integration: Using computer use to operate the screens of systems that will never get an API. The largest architecture requirement in the newest YC batch, and a proven pattern in the customer corpus.
Updated 2026-07-22 (Reference) Glossary: One-line-per-term summary across the whole wiki.
New 2026-07-22 (Concept) Consumption Pricing: The Claude Developer Platform bills by tokens consumed, not seats sold. A builder's bill scales with how much their product actually runs, which aligns the platform with usage that works.
New 2026-07-22 (Concept) Memory Stores: Workspace-scoped persistent memory for Managed Agents: a filesystem mounted into every session, versioned, with an audit trail, so what an agent learns survives the run.
New 2026-07-22 (Concept) Outcomes: The Managed Agents primitive for defining done: you supply a rubric, and the platform runs an iterate, grade, revise loop with a separate grader until the work passes.
New 2026-07-22 (Concept) Prompt Caching: Cache reads cost a fraction of base input price, but the cache is a strict prefix match: one changed byte invalidates everything after it. Stable content goes first.
New 2026-07-22 (Concept) Scheduled Deployments: Cron for agents on Managed Agents: a schedule fires a fresh session on the clock, every firing leaves a run record, and the whole thing can be paused or archived.
New 2026-07-22 (Concept) The Batch API: 50 percent off input and output tokens for anything that can wait up to 24 hours. The discount stacks with prompt caching.
New 2026-07-22 (Concept) The Effort Dial and Model Splitting: Spend frontier-model tokens on judgment and small-model tokens on commodity steps, and turn the effort parameter down where deep thinking is wasted.
New 2026-07-22 (Concept) The Human Gate: An approval step the platform enforces before an agent's action commits. A prompt can ask for approval; a permission policy guarantees it.
New 2026-07-22 (Concept) Vaults: The Managed Agents credential store: secrets never enter the sandbox, they are substituted into outbound requests at egress, so a prompt injection has nothing to steal.
New 2026-07-22 (Concept) Agent Once, Session Every Run: The Managed Agents usage rule: create a versioned agent config once, then create a session per run. Creating agents in the request path is the documented anti-pattern.
New 2026-07-22 (Concept) The Four Ways to Build: Manual loop, Tool Runner, Agent SDK, Managed Agents: the four paths onto the platform, separated by who supplies the harness and who supplies the deployment.
New 2026-07-22 (Concept) The Deployment: Everything an agent runs on: sandboxing, session persistence, tenant isolation, secrets, and capacity. The infrastructure half of the agent question.
New 2026-07-22 (Concept) The Harness: The software loop around the model: tool dispatch, context management, retries, compaction. The half of an agent that is code, not model.
Updated 2026-07-22 (Case Study) The Startup Cut: Anthropic tags 29 customer stories as startups, and 21 of the 256 are Y Combinator companies at every stage from current batch to exit.
Updated 2026-07-22 (Case Study) Strength of Evidence: Not all published metrics are equally defensible. A tiering of the corpus by method, plus the specific pages that contradict themselves.
Updated 2026-07-22 (Case Study) The Index: Every published Anthropic customer story in one searchable table: 256 rows, each linked to the official first-party page.
Removed 2026-07-22 (Concept) Sample Concept: A one-line definition of the sample concept. Quotable, plain, no rhetorical flourishes.
Updated 2026-07-22 (Reference) Glossary: One-line-per-term summary across the whole wiki.
New 2026-07-22 (Case Study) The Index: Every published Anthropic customer story in one searchable table: 256 rows, each linked to the official first-party page.
New 2026-07-21 (Case Study) The Startup Cut: Anthropic tags 29 customer stories as startups, and 21 of the 256 are Y Combinator companies at every stage from current batch to exit.
New 2026-07-21 (Case Study) Strength of Evidence: Not all published metrics are equally defensible. A tiering of the corpus by method, plus the specific pages that contradict themselves.
New 2026-07-21 (Concept) Sample Concept: A one-line definition of the sample concept. Quotable, plain, no rhetorical flourishes.
New 2026-07-21 (Reference) Glossary: One-line-per-term summary across the whole wiki.
New 2026-07-21 (Reference) Graphic Style: Visual conventions for illustrations and diagrams on this wiki. The default is comic-strip article summaries in the Supersuit Up family style.
New 2026-07-21 (Reference) Voice Rules: Writing constraints every page is held to. Edit per wiki.