.png)
3 months ago
1
for software engineers
The amount of noise and hype around generative AI is huge. Is it going to replace software engineers? Will teenagers be vibecoding financial processing systems while watching TikTok videos? Are we going to be reduced to tidying up reams of sloppy AI code?
Or is it just another tool that is helpful when used skillfully and judiciously? If so, where to start? How to cut through all the noise about models, agents, hallucinations, distillation, quantisation and so on? What are the ethical, productivity and other considerations in adopting such a tool?
NO HYPE AI is my attempt to cut out the hype and provide a level-headed overview of generative AI tools for software engineers.
Overview
A very short crash course to introduce the terminology
If you haven't spent much time diving into generative AI, read this
intro to get familiar with high level concepts and jargon.
Tools
Generative AI tools for software development
Choose an IDE, IDE extension, or a CLI tool. Your choice of cloud
or local LLM can often be paired with the tools.
How to use
Practical tips and advice for using generative AI
Context engineering, structuring the interaction with LLM, iterating effectively. Guidelines, how-tos, recommendations.
Models
Cloud and local model providers and models
Cloud and local models. How model providers, model developers, model routers and model repositories fit together.
Benchmarks
Performance comparisons and evaluations for models and agents
Links to various leaderboards and comparisons along with a description of caveats in model assessment. Some discussion of findings on the effects on developer productivity.
Why NOT use AI?
Reasons not to use AI (yet?) for software development
There is a range of ethical, legal, security, and productivity questions to consider. It's also a rapidly changing space with no established best practices.
Glossary
Key terms and concepts explained
Short definitions of all the genAI jargon. Old school, I know—but free of hallucinations.
Papers
Some papers about LLMs, agents, and their effects
Not required reading in order to use LLMs and agents, but if you like going to the sources, here are some.
Vibecoding reports
Substantive reports of success with vibecoding
As yet unclear whether vibecoding is viable or desirable, but
these are at least interesting data points on what can be achieved
with only generated code.
