DVx Signal/AI at Work/Issue 008

Interactive ChatGPT, dots, meeting notes and two Claude models

Try new ways to work in ChatGPT, then choose the Claude model that fits the task.

DVx Research4 min read

ChatGPT can answer with a tool you can use

October 7 rollout · checked October 8Official source ↗
OpenAI's Intelligent UI bill splitter with editable guest and item fields inside a ChatGPT-style response

OpenAI’s live bill-splitter example has editable fields inside the answer, rather than a pasted list. View official source ↗

What it does

GPT-6 Intelligent UI can put buttons, forms, charts, and small interactive tools inside a ChatGPT answer. OpenAI says a component library and compiler build the interface as the answer streams.

Who gets it

The Chat tab rollout starts October 7 for Plus, Pro, Business, and Enterprise globally; Free and Go start October 8. Enterprise admins control access. This release does not change the models in ChatGPT Work or Codex.

Your move

Ask ChatGPT to turn one calculation you already do into an editable tool. Change an input and check the result against your own calculation before using it.

A dot can notice work in connected apps

September 29 announcement · gradual rollout · checked October 8Official source ↗
OpenAI launch example of a dot finding an iOS checklist problem and preparing a fix for a person to review

OpenAI’s illustrative demo shows a dot preparing a fix and asking its user to review it. View official source ↗

What it does

Dots are persistent ChatGPT agents with their own cloud computers. They can look through apps you connect and bring work to you. Background discovery uses read-only app tools; sending or changing content involves separate permissions and review.

Who gets it

One dot is included on eligible Pro and Business Premium plans, with a deeper-work allowance. OpenAI’s access guide excludes the EEA, UK, and Switzerland from Pro at launch; Enterprise, Edu, and Healthcare require an admin-enabled beta. Access may take days.

Your move

Connect only the apps needed for one narrow job, such as watching for a missed project deadline. Review the app permissions and the dot’s activity before letting it handle more.

Sonnet 5.5 is a sensible first try for routine work

September 28 launch · October 7 cache change · checked October 8Official source ↗
MeasureSonnet 5.5
API / 1M$2 in / $10 out
Task costUp to 30% less*
Cache read$0.10, was $0.20
*Anthropic’s test, not an independent result. API rates, not a change to your Claude subscription price. Cache rate changed October 7. View official source ↗
What it does

Anthropic positions Sonnet 5.5 for well-scoped writing, spreadsheets, slides, and bug fixes. It reports faster output and fewer tokens per task than Sonnet 5.

Who gets it

Sonnet 5.5 is available on Anthropic’s platforms. Its API input and output rates match Sonnet 5; Anthropic halved its cache-read rate on October 7. Subscription usage limits still depend on your plan.

Your move

Give it a real document or spreadsheet you can verify. Compare the corrections needed against your current model before switching your default.

Meeting notes can become a task you can hand off

September 29 beta · macOS desktop · checked October 8Official source ↗
OpenAI Meetings release image showing a meeting recap, an action item, and an ask to ChatGPT

OpenAI’s release image shows a recap and an action item handed to ChatGPT. View official source ↗

What it does

The Meetings plugin records microphone and system audio, then saves a private recap and suggested actions in ChatGPT Space. It does not put a bot in the call.

Who gets it

It is a macOS desktop beta for Pro and Business. Enterprise access is limited to an alpha group. Business admins must enable both Space and Meetings.

Your move

Tell everyone you are taking notes and get consent first. Try it on one internal meeting, correct the recap, and review each action before sharing the page. OpenAI says the audio is deleted after notes are ready.

Use Haiku 5.5 for short, repeated work

October 7 launch · checked October 8Official source ↗
Work to testStart with
Sort or summarize many short itemsHaiku 5.5
Polish a client-ready documentSonnet 5.5
A DVx starting point based on Anthropic’s model guidance. Compare the results on your own tasks. View official source ↗
What it does

Haiku 5.5 is Anthropic’s faster small model for quick, repeated tasks such as summaries and classification. Anthropic reports about 75% lower cost per task than Haiku 4.5 on average.

Who gets it

Anthropic says it is available on all its platforms. In Claude, model access and usage still depend on your plan. The lower task-cost claim is Anthropic’s average, not a subscription discount.

Your move

Use the same batch of short items with Haiku and Sonnet. Check mistakes and time saved. Keep Sonnet for the final artifact when its extra care changes the result.

Also since the last issue

Useful updates

Gemini 4 Argon · limited access

Google announced it September 30 with a 1M-token output limit, but only selected cyber defenders have access now. Wait for a general rollout before planning a switch.

28 days of shipping · ongoing

The Developer Community’s live tracker changes daily. Check your current usage limits in the product instead of assuming a reset or speed change applies to your account.

GPT-6.1 Sol · access

OpenAI released it September 29 in ChatGPT Work and Codex for paid plans. It is separate from the October 7 GPT-6 Chat rollout.

ChatGPT Space · team access

OpenAI’s September 29 release gives Pro, Business, and Enterprise a place for shared pages and agent work. Meetings notes can live there; check your workspace settings.

15-minute try

Make one answer interactive

In ChatGPT Chat, ask for a small editable calculator using numbers you already know. Change one input, check the new result, and keep it only if the answer is easier to use than a plain table.

Before you upgrade

Three checks

Do you already have access? Is there a named task it improves? Can you show a better result after one test? If any answer is no, wait before paying for a higher plan.