AI tools for newsrooms.

# Vibe coding quick start checklist

Pre-flight checklist for AI-assisted coding sessions. Print this out or keep it open during your first few projects.

---

## Before you start

- [ ] **Define a specific goal** — Write down exactly what "done" looks like in one sentence
- [ ] **Gather your data** — Have sample files ready, know your data format
- [ ] **Choose your LLM** — Pick one and stick with it for this session
- [ ] **Create project folder** — Clear name, organized structure
- [ ] **Start a changelog** — Use the CHANGELOG-TEMPLATE.md file
- [ ] **Test your environment** — Can you run "Hello World" in your language of choice?
- [ ] **Set a time limit** — Protect yourself from rabbit holes

---

## During the session

### Prompting best practices

- [ ] Start with the big picture, then drill into details
- [ ] Include your environment info (OS, language version, key libraries)
- [ ] Provide sample input/output when possible
- [ ] Ask for code in chunks, not all at once
- [ ] When stuck, ask "What might be causing this?" not just "Fix this"

### When code breaks

- [ ] Copy the **full error message**, not just the last line
- [ ] Paste it back to the AI with context: "I ran this code and got this error"
- [ ] If same error 3+ times, step back and explain the bigger goal
- [ ] Check if the error is your environment, not the code

### Red flags

- [ ] Code getting too complex? Split into smaller scripts
- [ ] Scope creeping? Note the new idea for later, stay focused
- [ ] Going in circles? Take a break, return with fresh eyes
- [ ] AI giving same wrong answer? Try rephrasing or different LLM

---

## Before you stop

- [ ] **Test with real data** — Not just the sample
- [ ] **Document what you built** — Update the changelog
- [ ] **Note what's unfinished** — Clear next steps for future you
- [ ] **Save working state** — Commit to git or backup the folder
- [ ] **Clean up** — Remove test files, add comments to unclear code

---

## Quick reference

### When to use which language

| Task | Language |
|------|----------|
| Data cleaning/analysis | Python |
| Statistical analysis | R |
| Interactive web visualization | JavaScript |
| File automation | Bash/Python |
| Database queries | SQL |

### Current LLM starting points for coding

> Verified August 15, 2026. Model names, availability, and pricing change quickly; check the linked provider documentation before production use.

| Need | Recommendation |
|------|----------------|
| Maximum-capability Anthropic work | Claude Fable 5 |
| Advanced Anthropic coding and reasoning | Claude Opus 5 |
| Balanced Anthropic coding | Claude Sonnet 5 |
| Current Google coding and agents | Gemini 3.7 Flash |
| OpenAI coding tasks | Codex (GPT-5.6 Sol) |
| Balanced OpenAI agents | GPT-5.6 Terra |
| High-throughput OpenAI tasks | GPT-5.6 Luna |
| Highest hosted Qwen evaluation | Qwen 3.8 Max preview |
| Stable hosted Qwen agents | Qwen 3.7 Plus |
| Open-weight, long-horizon agents | GLM-5.2 |

---

*From [Center for Cooperative Media](https://centerforcooperativemedia.org) AI Tools for Newsrooms*

# LLM comparison chart for journalism projects

> Verified August 15, 2026. The AI landscape changes fast, so verify current pricing, availability, privacy terms, and capabilities.

---

## At a glance

| Model | Best for | Speed | Cost | Free tier? |
|-------|----------|-------|------|------------|
| Claude Fable 5 | Maximum-capability long-running agents | Slower | $$$ | Check current plan |
| Claude Opus 5 | Complex agentic coding and writing | Slower | $$$ | Check current plan |
| Claude Sonnet 5 | All-around tasks, best balance | Fast | $$ | Yes |
| Claude Haiku 4.5 | Fast, cheap, high-volume tasks | Fastest | $ | Limited |
| Gemini 3.1 Pro (preview) | Difficult multimodal reasoning | Slower | $$ | Check current plan |
| Gemini 3.7 Flash | Current coding and agent workhorse | Fast | $ | Check current plan |
| Gemini 3.6 Flash | Previous stable Flash tier | Fast | $ | Check current plan |
| Gemini 3.5 Flash | Earlier stable Flash tier | Fast | $$ | Check current plan |
| Codex (GPT-5.6 Sol) | OpenAI coding tasks, long-horizon refactors | Fast | $$ | Limited |
| GPT-5.6 Sol | Deep research, complex reasoning | Slower | $$$ | Limited |
| GPT-5.6 Terra | Balanced agents, coding, and knowledge work | Fast | $$ | Check current plan |
| GPT-5.6 Luna | High-throughput routine work | Fastest | $ | Check current plan |
| Qwen 3.8 Max (preview) | Highest hosted Qwen evaluation tier | Varies | $$ | Check current plan |
| Qwen 3.7 Plus | Stable hosted long-context and multimodal agents | Varies | $$ | Check current plan |
| GLM-5.2 (open) | Long-horizon open-weight agents | Varies | Hosting* | Self-host |
| DeepSeek V4 Pro/Flash (open) | Long-context coding and tools | Varies | Hosting/API* | Self-host |
| Qwen3.6-35B-A3B (open) | Efficient local agentic coding | Varies | Hosting* | Self-host |
| Kimi K3 (open) | Long-horizon coding and multimodal knowledge work | Varies | Hosting* | Self-host |
| MiniMax M3 | Open-weight multimodal coding and agents | Varies | Hosting* | Self-host or API |
| Mistral Medium 3.5 / Small 4 (open) | Private multimodal and efficient agents | Varies | Hosting/API* | Self-host |
| Step-3.7-Flash / MiMo-V2-Flash / Hy3 (open) | Efficient agent and coding evaluations | Varies | Hosting* | Self-host |
| Command A+ / Nemotron 3 Ultra / Llama 4 (open) | Enterprise orchestration and deployment options | Varies | Hosting* | Self-host |

\* Open-weight models are free to use, but you pay for hosting, hardware, or inference infrastructure.

---

*From [Center for Cooperative Media](https://centerforcooperativemedia.org) AI Tools for Newsrooms*

# Project Name — Changelog

> Use this template to document each vibe coding session.

---

## [YYYY-MM-DD] Session Title

### Goal
<!-- What you set out to accomplish this session -->

### What worked
-

### What broke
- **Error:**
- **Fix:**

### Code changes
- `filename.py` — Description

### Next steps
- [ ]

---

*Template from [Center for Cooperative Media](https://centerforcooperativemedia.org) AI Tools for Newsrooms*

# Lessons learned — [Project Name]

> Use this template to document what you learned from a completed or abandoned project.

**Project:** [Name]
**Date range:** [Start] – [End]
**Status:** [Completed / Abandoned / On hold]

---

## Project summary
<!-- One paragraph: What was this? What problem did it solve? -->

## What worked
### Technical wins
-

### Process wins
-

## What didn't work
### Critical failures
-

### Technical debt
-

## The real problem
<!-- What was *actually* needed vs. what you initially set out to build? -->

## What I'd tell past me
1.
2.
3.

## Recommendations
### If continuing this project
1.

### If starting fresh
-

---

*Template from [Center for Cooperative Media](https://centerforcooperativemedia.org) AI Tools for Newsrooms*