alexjameson

I'm a technical writer and I use agents a lot for my work and personal projects. I also created a basic open-source preset for agents to create end-user documentation.

During the preset testing, I created dozens of READMEs using agents. Although I can't say that any of them were perfect for me, most of them were good enough.

README is a very special kind of document. For small projects, it's often the only one that potential users will read. The README should explain what the product is, what problems it solves, how to get started, provide some examples, and explain the restrictions. It should cover all these topics in a concise way so that even readers who aren't used to read long texts can easily read through it. It's definitely up to human to make it clear and easy to understand.

It's right for small projects that are created by people who know how to code in geberal. But there are also two categories where this might not be true:

1. Simple utilities that a vlogger or a sales guy created to automate routine. Such utilities are created in a hurry, deployed with prodiding private key to an agents, and so on. No attention to security was payed.

2. New generation of enterprise-grade slop that is produced by a relatively large group of people that don't review their work properly due to insane TTM and amounts of code.

Yes, their code might be secure in terms of simple techniques (agents help a lot of course), but in such code bases it's extremely hard to find out what should be fixed, how to fix it without breaking something else, and how to prevent something similar from happening in the future.

As a tech writer myself, I often hear from other tech writers around the world that they find themselves as editors for llm-written slop. The main problem is that even before llns, almost no senior manager understood what the TWs actually do. As a result, there is a strong desire to get rid of the editor team.(see the case of Snowflake when a team of 70 documentarians was let go in a single day).

The company I work for also encourages us to use agents as much as possible. Now, I spend more time in the terminal than in VS Code. However, the situation is different for those who work in regulated areas.

Most docs are here https://www.kimi.com/code/docs/en/kimi-code-cli/customizatio...

Not sure about the feature you mentioned but one can see in their app that they use IMF, World Bank, SEC EDGAR, and Yahoo Finance.

Not necessarily. I had almost no idea what I was doing when I started using Pi. The goal was to learn harness engineering and understand how it feels like thinking about things that are provided in other harnesses by default.

It took time to learn all of these before I managed to do something similarly useful as OpenCode, and now I am ready to use some sane defaults for things that I really know I need in Pi.

OK, got it. But it's important to remember that avoiding restrictions gets harder and harder every year. You have to juggle several protocols, have backup proxies, think about TLS nuances, keep in mind that you are being tracked after all. Lots of casual users just gave up. It's not like we find a solution and it just works — one has to adapt to new restrictive measures continuously.

I just don't understand why some other people in such discussions still think that its only about children and weigh pro and cons when they have an actual example of all these things being implemented.

Let me tell you how it has been in Russia during last 15 years. WE saw targeted blocks that apply after cease and desist letters. Later they learned to block outgoing OpenVPN and wireguard for everybody except firms who applied to a special registry. Then they learned to cut Vless and Trojan and blocked all sites behind freaking Cloudflare due to ECH v3.0 enforced. Here also go proxies (specifically for Telegram'm MTProto) and other stuff like Quic.

The point is if they start — they can proceed way further than you can imagine now.