Calm Capacity Brief | Issue 001 · AI Capacity for Nonprofits

Latoya Robinson

28d (edited) • Templates & Workflows

Calm Capacity Brief | Issue 001

Practical AI. Stronger systems. Room for the mission.

The Calm Infrastructure Lab | LUR Growth

Issue 001 | September 4, 2026

This week: make one report easier to finish

There is a lot of AI news right now. You do not need to keep up with every release to make useful progress.

For this first Calm Capacity Brief, I want to connect the updates to something familiar: the report that pulls you into three spreadsheets, two email threads, and a search for the latest numbers.

The starting point is a clear process, trustworthy information, and a person who reviews the result.

Three updates worth understanding

1. Claude: new models, the same need for clear boundaries

On September 1, Anthropic announced Claude Fable 5.1 and Mythos 5.1. Fable 5.1 is generally available; Mythos 5.1 is restricted to trusted access programs. Anthropic also described new enterprise safeguards with a phased rollout planned for later this fall. Those safeguards should not be assumed to be active in your account today. Read Anthropic's announcement.

My practical take: if you use Claude or Claude Code, test a familiar task with fictional data before changing a working process. Check the result against the source, including what the model leaves out. A stronger model still needs a defined job and a reviewer.

2. ChatGPT and Codex: more capability, staged access

OpenAI's update for August 31 through September 4 describes the initial GPT-6 Astra rollout for enterprises in its Trusted Access Program, with broader access expected afterward. It is not a promise that every member can select it today. Read the official weekly update.

My practical take: do not rebuild your process around an announcement. If a new model becomes available, compare it with your current tool on one repeatable task. Count review time and corrections, not just how quickly it produces a first draft.

3. Perplexity: a more explicit pause before actions

Perplexity's August 24 update describes an “Always ask” control for connector tools. It also notes that recurring runs inherit permissions granted at the thread level. This is a recent update, not a release from this week. Read the connector approval update.

My practical take: start with drafts. Check whether the tool can send messages or change records, and keep those actions behind human approval. Review standing permissions before repeating a workflow.

Also on the radar

Perplexity announced Hybrid Compute on Mac on September 1. It combines cloud work with local processing and a privacy gate. This is worth watching, but a vendor privacy feature is not permission to upload confidential records. Have your organization's technology or privacy lead review the actual configuration first. Read Perplexity's explanation.

Claude Design is real, but it is not a new launch this week. Introduced April 17, it supports visual work such as slides, one-pagers, and prototypes, with a handoff to Claude Code. A useful first experiment is a board update made from already approved, nonconfidential content. Check every number and label before sharing. Read the Claude Design introduction.

Your 15-minute capacity practice

Choose one recurring report and write down:

What decision or deadline it supports.
Where the approved numbers live.
Who owns the draft and who reviews it.
What information must stay out of the AI tool.
What happens when a number is missing or two sources disagree.

Try this instruction using fictional information:

Turn the approved summary below into a short report draft. Use only the supplied facts. Do not invent outcomes, people, dates, or explanations. Mark missing information as NEEDS REVIEW. For each factual claim, identify the source label that supports it. Separate the draft from questions for the reviewer. Do not send or publish anything.

Then compare the draft with the source. If the checking takes longer than the old process, that is useful information. Simplify the workflow before adding more automation.

Resources you can use today

The workflow worksheet accompanying this issue gives you a place to capture the five decisions above and a small test log.

LUR Growth also has a free AI use policy template and sample capacity diagnostic. The policy is a starting point for organizational review, not a substitute for your own approved rules.

If you are returning to the August 21 session, the replay is here. The community replay post points to minute 37 as the start of the live build. You do not need to watch the whole session before trying one small step.

Which recurring report would you like to make easier first?

Latoya

0 comments

Calm Capacity Brief | Issue 001