# Monday playbook: Agent Hours

Resource for the AI Deep Dive on 2 November 2026 · Momo Maximilian Feichtinger · zukunftbilden.eu/deepdive

Five steps for your first agent hour, a template for the brief and six plays with prompts. Prices and availability as of 21 October 2026.

## Five steps to your first agent hour

1. **Pick one task:** Something recurring that costs you two hours a week and has a clear result.
2. **Pick the tool:** Claude Code or Cowork for files and builds, Codex cloud for code, Dots for ongoing watch tasks.
3. **Write the brief:** Goal, inputs, what done looks like, limits, how to report.
4. **Give little access:** One folder, its own account, one budget. Nothing the agent could break.
5. **Check in the morning:** Read the report, check one result in depth, improve the brief.

## The brief template

Copy it, fill the angle brackets, paste it into Claude Code, Codex, Cowork or ChatGPT agent mode.

```text
GOAL      Prepare <the result> for <who> by <when>.
INPUTS    Use only: <folders, files, links>.
DONE      Finished when: <three checkable criteria>.
LIMITS    Do not send, buy, delete or publish anything.
          Max <n> hours, max <budget>. Ask if unsure.
METHOD    Write a five-line plan first, then work.
          Check your result against DONE before you stop.
REPORT    One page: what you did, what is open,
          what you are unsure about, where the files are.
```

## Six plays with prompts

### 1. An explainer animation from one prompt

**Tool:** Claude Code with Opus 5.5 (from Claude Pro, 20 dollars a month) or Claude Motion (beta on Team)

Opus 5.5 writes animations as code, with no video generator. A clear message and a style hint are enough for a one-minute explainer. You can change the result sentence by sentence.

**Prompt:**

```text
Create a 60-second explainer animation as HTML with SVG and GSAP: “How a heat pump works”. Six scenes, calm motion, English labels, one accent colour. Render it as a 1080p MP4 and put the script and source files in ./film.
```

**Note:** Use music and voices only from clearly licensed sources. Label AI-generated content where Article 50 of the AI Act requires it.

### 2. An interactive learning lab for class

**Tool:** Claude (also Free) with artifacts or Claude Code, ChatGPT with Canvas

In October a single teacher built a lens lab in one run: sliders, ray diagram, exercises. Students learn by trying, and the teacher checks every value beforehand.

**Prompt:**

```text
Build an interactive learning lab for 14-year-olds on converging lenses. Sliders for focal length and object distance, a drawn ray diagram, three exercises with instant feedback that give a hint and never the answer. One single HTML file that works offline.
```

**Note:** Spot-check the physics values yourself before the lab goes to class.

### 3. The night shift for code

**Tool:** Codex cloud with GPT-6.1 Sol (from ChatGPT Plus) or Claude Code with a scheduled task

Maintenance work such as updating dependencies, adding tests or fixing small bugs runs overnight in the cloud. In the morning a pull request is waiting for review.

**Prompt:**

```text
Work through the list in TODO-maintenance.md, one item per commit. Do not change public interfaces. Run all tests after each item. If a test fails twice, leave the item open and write why. At the end: one pull request with a summary, no merge.
```

**Note:** The agent gets the repository only, no production keys. Merge and deploy stay with you.

### 4. Answer inquiries, as drafts

**Tool:** Claude with a saved skill, later a scheduled task; Gmail or Outlook connection

Incoming inquiries are read, sorted and prepared as reply drafts. Only you send. That way you get to know the agent with no risk for customers.

**Prompt:**

```text
Read the inquiries in the folder “Inbox” from the last 24 hours. Assign each to a category (offer, appointment, question, other). Write a reply draft for each in my tone from example-replies.md and save it as a draft. Send nothing. At the end, a list: sender, category, open question for me.
```

**Note:** Process personal data only on a plan with suitable data protection terms.

### 5. A radio play with several voices

**Tool:** Opus 5.5 as director, voices via Gemini TTS, ElevenLabs v4 or GPT-Live (German voices since 15 Oct)

A chapter of my novel became a radio play with nine voices and music, and the voices cost 1.60 dollars. Opus writes the script, casts the roles and drives the speech output through the API.

**Prompt:**

```text
Turn chapter-1.md into a radio play of about ten minutes. Write a script with a narrator and at most four characters, choose fitting voices, generate every line separately through the TTS API and mix everything with pauses and quiet music into one MP3. Add a cost breakdown.
```

**Note:** Do not imitate real people. Clone voices only with explicit consent.

### 6. A dashboard from your numbers

**Tool:** Claude Dashboards (beta since 8 Oct) or Claude Code with a CSV file

A table of sales, sign-ups or stock levels becomes an overview that updates itself with new data. The agent explains every metric in one sentence.

**Prompt:**

```text
Here is sales-2026.csv. Build a dashboard with the five metrics that matter most for a small company, with a month-on-month comparison and a warning when a value moves by more than 20 percent. Explain each metric in one sentence. Recalculate every total a second time as a check.
```

## Prices and plans

Vendor list prices on 21 October 2026 in US dollars per month unless stated otherwise. In Austria some vendors charge in euros including VAT. Check the vendor page before you buy.

| Plan | Price | For whom | Note |
| --- | --- | --- | --- |
| Claude Pro | $20 | One person getting started: documents, slides, first agent hours | Five-hour and weekly limits |
| Claude Max 5x / 20x | $100 / $200 | Anyone building with Claude Code every day | Monthly API credit included since 7 Oct |
| Claude Team | $25 per person | Small teams with shared projects | Minimum two seats, Motion in beta |
| ChatGPT Plus | $20 | Getting started with Codex, agent mode and images | Codex and chat share one allowance |
| ChatGPT Pro 100 / 200 / 500 | $100 / $200 / $500 | Heavy Astra use and long Codex runs | New $200 subscribers since 29 Sep get half the allowance; Dots are not on Pro in the EU |
| ChatGPT Business | $25 per person | Companies that need a data processing agreement | Dots in the EU with Business Premium or Enterprise |
| Google AI Pro | €21.99 | Bundling Gemini, Workspace and storage | Austrian price including VAT |
| Mistral Le Chat Pro | $14.99 | European option for chat and research | excluding VAT |

## API prices per million tokens

| Model | Since | Input | Output | Note |
| --- | --- | --- | --- | --- |
| Claude Opus 5.5 | 22 Sep | $4 | $20 | Anthropic's top model |
| GPT-6.1 Sol | 29 Sep | $2 | $10 | near Astra level at a fifth of the price |
| Gemini 4 Argon | 13 Oct | $2 | $10 | introductory price until 30 Nov, then $4 / $20 |
| Claude Sonnet 5.5 | 28 Sep | $2 | $10 | Anthropic's everyday model |
| Claude Haiku 5.5 | 7 Oct | $0.10 | $0.50 | for short, frequent tasks |
| GPT-6 Luna | 22 Sep | $0.10 | $0.50 | for large volumes of simple tasks |
| DeepSeek V4.1 Flash | 17 Oct | $0.30 | $1.20 | open weights under the MIT licence |
| Mistral Large 4 | 6 Oct | $1.36 | $4.18 | European, preview |
| GPT-6 Astra | for comparison | $10 | $50 | the most expensive model in this table |

## Dates

- **Mon, 2 Nov · Deep Dive: Agent Hours:** 7:30 pm, penthouse at Magazinstraße 4 in Salzburg and online. The ARC-AGI-3 Kaggle competition closes the same day.
- **Tue, 3 Nov · EU consultation on copyright and AI:** Last day for feedback to the European Commission.
- **Mon, 30 Nov · Gemini 4 Argon: introductory price ends:** The API price then doubles to 4 and 20 dollars.
- **Thu, 3 Dec · EU KIDS Act: feedback deadline:** Age limits for social media and parental controls for chatbots under 13. Relevant for schools.
- **Mon, 7 Dec · Next Deep Dive:** 7:30 pm, as always on the first Monday of the month.

---

Prices and availability as of 21 October 2026. Check the vendor page before you buy. The timeline with all sources is at https://zukunftbilden.eu/en/deepdive/resources/deepdive-2026-11-02
