Skip to content

Commit 16f6661

Browse files
samithoughtbotsamithoughtbot
committed
Add AI development tools guide
This guide compiles Thoughtbot's collective experience with AI tools in development to save time and improve knowledge sharing. It summarizes: - Frequency of AI usage - AI models and tools being used - Where AI is used (editor, browser, terminal, etc.) - Tasks AI supports (coding, debugging, testing, code review, planning) - Prompting strategies employed - Agentic AI usage and associated tasks - Tools used by task How it works: 1. **Survey and gather input** – Once a quarter, the team provides input on which AI tools they are using, how they are using them, and what has or hasn’t worked. 2. **Create a shared guide** – Survey responses are collected here to give visibility into what people are actually using, allowing team members to learn from others’ experiences and decide what to experiment with themselves. This outcome is the result of the **Fusion-based AI task force**. - Initial creation of the AI task force: [How to get the best out of AI task force](https://hub.thoughtbot.com/messages/how-to-get-the-best-out-of-ai-task-force) - Link to the survey: [The AI Development Survey](https://hub.thoughtbot.com/messages/the-ai-development-survey) The guide is a living resource, updated regularly with survey input, to help team members learn from each other’s experiences rather than starting from scratch. Co-authored-by: Jose.blanco@thoughtbot.com Co-authored-by: rakesh@thoughtbot.com
1 parent 4dac788 commit 16f6661

1 file changed

Lines changed: 133 additions & 0 deletions

File tree

‎artificial-intelligence/README.md‎

Lines changed: 133 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,133 @@
1+
This guide exists to save time by showing which AI tools are being used and how, so everyone can learn from collective experience rather than starting from scratch.
2+
3+
AI is evolving quickly, and people at thoughtbot are experimenting with different tools in their work. Instead of everyone spending time evaluating tools on their own, this guide brings together our shared experiences so we can learn from each other and make faster, better-informed choices.
4+
5+
This makes the guide a living resource, updated regularly to reflect how our use of AI is changing. It is not a fixed set of rules, but a practical reference that saves time, helps people get started more quickly, and ensures our decisions are built on collective experience.
6+
7+
Here’s how it works:
8+
• Survey and gather input: Once a quarter, we ask everyone what AI tools they are using, how they are using them, and what has or hasn’t worked.
9+
• Create a shared guide: The survey responses are collected here to give visibility into what people are actually using. This lets everyone see the range of tools in play, learn from others’ experiences, and decide what they’d like to experiment with themselves.
10+
11+
# AI in Development Survey Q3 2025 – Summary
12+
13+
## 1. Frequency of AI Usage
14+
| Frequency | Count |
15+
|-----------|-------|
16+
| Daily | 12 |
17+
| A few times a week | 2 |
18+
| Occasionally | 3 |
19+
| Rarely | 1 |
20+
| Never | 1 |
21+
22+
> **Insight:** Most developers in the survey use AI daily in their work.
23+
24+
---
25+
26+
## 2. AI Models Used
27+
| AI Model | Count |
28+
|----------|-------|
29+
| OpenAI (GPT-3, GPT-4, DALL-E) | 12 |
30+
| Anthropic (Claude) | 11 |
31+
| Google AI (Gemini, LaMDA, PaLM) | 2 |
32+
| Open Source Models | 1 |
33+
| CodeRabbit | 1 |
34+
35+
> **Insight:** OpenAI and Anthropic are the dominant AI models among respondents.
36+
37+
---
38+
39+
## 3. AI Tools / Software Used
40+
| Tool / Software | Count |
41+
|----------------|-------|
42+
| ChatGPT | 12 |
43+
| GitHub Copilot | 7 |
44+
| Claude Code | 6 |
45+
| Cursor | 3 |
46+
| Roo Code | 1 |
47+
| gptel in Emacs | 1 |
48+
| Other | 4 |
49+
50+
> **Insight:** ChatGPT is the most commonly used AI tool, followed by Copilot and Claude Code.
51+
52+
---
53+
54+
## 4. Where AI is Used
55+
| Location | Count |
56+
|---------|-------|
57+
| Inside the code editor | 12 |
58+
| In the browser | 8 |
59+
| In the terminal | 5 |
60+
| Native apps | 2 |
61+
| ChatGPT software/app | 2 |
62+
| Sometimes in desktop apps | 1 |
63+
64+
> **Insight:** Developers mostly use AI integrated into their code editor, with secondary use in browsers or terminals.
65+
66+
---
67+
68+
## 5. Tasks AI is Used For
69+
| Task | Count |
70+
|------|-------|
71+
| Writing code | 14 |
72+
| Debugging code | 11 |
73+
| Writing tests | 8 |
74+
| Explaining unfamiliar code | 11 |
75+
| Planning features / architecture | 8 |
76+
| Writing commit messages | 2 |
77+
| Generating documentation | 3 |
78+
| Reminders / recalling knowledge | 1 |
79+
| Autocomplete | 1 |
80+
| Finding best practices / options | 1 |
81+
82+
> **Insight:** Core programming tasks like writing code, debugging, and understanding code are the most common AI applications.
83+
84+
---
85+
86+
## 6. Prompting Strategies
87+
| Prompt Strategy | Count |
88+
|----------------|-------|
89+
| No specific strategy | 5 |
90+
| Zero-shot prompting | 4 |
91+
| Few-shot prompting | 5 |
92+
| Chain-of-thought prompting | 2 |
93+
| Persona-based prompting | 3 |
94+
| Constraint-based prompting | 5 |
95+
| Iterative prompting | 8 |
96+
| Asking clarifying questions | 7 |
97+
| Using templates / pre-defined prompts | 2 |
98+
| Giving hard constraints (e.g., tests must pass) | 1 |
99+
100+
> **Insight:** Iterative prompting and asking clarifying questions are highly used strategies. Many developers combine multiple strategies.
101+
102+
---
103+
104+
## 7. Agentic AI Usage
105+
| Usage | Count |
106+
|-------|-------|
107+
| Yes, frequently | 3 |
108+
| Yes, occasionally | 3 |
109+
| No, not yet | 10 |
110+
| Never | 1 |
111+
112+
**Agentic AI Tasks:**
113+
- Automated refactoring across multiple files
114+
- Generating end-to-end features from a high-level description
115+
- Automated testing and bug fixing workflows
116+
- Intelligent code migration/upgrades
117+
- Research and exploration of new technologies
118+
119+
> **Insight:** Most developers are **not yet using agentic AI**, but those who do focus on automation and feature generation.
120+
121+
---
122+
123+
## 8. AI Tools by Task
124+
| Task | Tools |
125+
|------|-------|
126+
| **Writing Code** | ChatGPT, Copilot, Claude, Cursor, Other |
127+
| **Debugging** | ChatGPT, Copilot, Claude, Cursor, Other |
128+
| **Writing Tests** | Copilot, Claude, Cursor, Other |
129+
| **Code Review** | Copilot, Claude, Other |
130+
| **Planning & Architecture** | ChatGPT, Claude, Other |
131+
| **Researching** | ChatGPT, Cursor, Other |
132+
133+
> **Insight:** ChatGPT and Copilot dominate coding and debugging, while Claude is frequently used for planning, testing, and research.

0 commit comments

Comments
 (0)