Lots of people making confident recommendations with the experience of a few weeks of exposure. I adopted OpenAI's Codex desktop app on day 1 of its release (a few days ago). I've done a few things with it. But honestly still figuring it out as I go. As anyone else in this space. Except for industry insiders, none of us have more than a few days experience with this. The models that got released last night are worse. Essentially everything you read about this is people rushing out articles on the back of at best a few hours of experience but otherwise just rehashing what's in the press releases.
My views on this:
- the models don't matter as much any more. You can pick Gemini, Claude, or Codex and get similarish results. Especially if you are not a rocket scientist and aren't trying to boil the oceans and are just trying to complete relatively straightforward development tasks. You probably don't need the most expensive models on "high". You get a new one every few weeks.
- The UIs and plumbing around the models matter a lot more. To orchestrate agentic loops around these models is a lot of plumbing and work. Claude Code and Codex are the two "it just works" type options in this space. Claude had all the momentum earlier than OpenAI. But Codex is getting a lot of glowing reviews now. If you are more of a tinkerer, there are a bazillion tool and model combinations to try out that don't meaningfully move beyond what those two do out of the box. Your mileage may vary. I'm mostly a codex guy. Because I'm lazy and the OpenAI Plus subscription is a very good deal.
- Google is good at models but not that great at tools. Yet. And they are worse at marketing them. If find their stuff very confusing.
- Any advice on what is "best" here has a half life measured in weeks. It can all be different in a few weeks.
I do like Antigravity a lot and I use it on a daily basis. Are you referring to that or something else? I do find their marketing weird though and seems like their internal teams are not yet fully aligned. (Jules, Antigravity, Gemini CLI, ...)
I've come across only a few of those. I'm not familiar with Antigravity. That's what I mean with poor marketing. I've tried out AI Studio, which is alright but not quite Codex level good as far as I can see.
Honestly this smells a bit like the situation they had a few years ago with a lot of different teams in Google inventing new chat and video call tools. All of which then flopped and got discontinued. Google exporting their own internal chaos. How many coding tools can you have in one company?!
OpenAI and Anthropic seem more focused currently. I mostly focus on OpenAI's codex because I don't want to maintain two subscriptions. Switching tools is a bit disruptive and not something I want to do every few days/weeks.
I treat any article on the subject as corporate output to lay cover to say “we know AI” when in fact, they are at the same place their customers are. Where we all are.
The answer to “which coding agent should we standardise on?” is really simple: Don’t.
The tradeoffs of the different models are complicated and difficult to wrap your head around, and if you have the resources to try a bunch and form a conclusion, next week new models will come out and change the equation in small but difficult-to-understand ways again.
The solution is to ask your engineers which models they like, get them access to as many of those as you can, and expect their preferences to change and price that in.
“But I don’t have the budget to buy subscriptions to all the models my engineers want! And there are compliance issues with some of them!”
Note that I didn’t say you have to get access to all of them: as many as you can. And try to push the envelope as much as possible. Get creative. Perhaps give engineers a $200/mo AI coding budget and let them pick from a selection of subscriptions. You’re going to have different engineers using different AI coding tools, and if you refuse to let them, your competitors will.
Maybe in the future “standardizing on one coding agent” will be a thing that makes sense. But that time is not now.
Lots of people making confident recommendations with the experience of a few weeks of exposure. I adopted OpenAI's Codex desktop app on day 1 of its release (a few days ago). I've done a few things with it. But honestly still figuring it out as I go. As anyone else in this space. Except for industry insiders, none of us have more than a few days experience with this. The models that got released last night are worse. Essentially everything you read about this is people rushing out articles on the back of at best a few hours of experience but otherwise just rehashing what's in the press releases.
My views on this:
- the models don't matter as much any more. You can pick Gemini, Claude, or Codex and get similarish results. Especially if you are not a rocket scientist and aren't trying to boil the oceans and are just trying to complete relatively straightforward development tasks. You probably don't need the most expensive models on "high". You get a new one every few weeks.
- The UIs and plumbing around the models matter a lot more. To orchestrate agentic loops around these models is a lot of plumbing and work. Claude Code and Codex are the two "it just works" type options in this space. Claude had all the momentum earlier than OpenAI. But Codex is getting a lot of glowing reviews now. If you are more of a tinkerer, there are a bazillion tool and model combinations to try out that don't meaningfully move beyond what those two do out of the box. Your mileage may vary. I'm mostly a codex guy. Because I'm lazy and the OpenAI Plus subscription is a very good deal.
- Google is good at models but not that great at tools. Yet. And they are worse at marketing them. If find their stuff very confusing.
- Any advice on what is "best" here has a half life measured in weeks. It can all be different in a few weeks.
I do like Antigravity a lot and I use it on a daily basis. Are you referring to that or something else? I do find their marketing weird though and seems like their internal teams are not yet fully aligned. (Jules, Antigravity, Gemini CLI, ...)
I've come across only a few of those. I'm not familiar with Antigravity. That's what I mean with poor marketing. I've tried out AI Studio, which is alright but not quite Codex level good as far as I can see.
Honestly this smells a bit like the situation they had a few years ago with a lot of different teams in Google inventing new chat and video call tools. All of which then flopped and got discontinued. Google exporting their own internal chaos. How many coding tools can you have in one company?!
OpenAI and Anthropic seem more focused currently. I mostly focus on OpenAI's codex because I don't want to maintain two subscriptions. Switching tools is a bit disruptive and not something I want to do every few days/weeks.
I treat any article on the subject as corporate output to lay cover to say “we know AI” when in fact, they are at the same place their customers are. Where we all are.
The answer to “which coding agent should we standardise on?” is really simple: Don’t.
The tradeoffs of the different models are complicated and difficult to wrap your head around, and if you have the resources to try a bunch and form a conclusion, next week new models will come out and change the equation in small but difficult-to-understand ways again.
The solution is to ask your engineers which models they like, get them access to as many of those as you can, and expect their preferences to change and price that in.
“But I don’t have the budget to buy subscriptions to all the models my engineers want! And there are compliance issues with some of them!”
Note that I didn’t say you have to get access to all of them: as many as you can. And try to push the envelope as much as possible. Get creative. Perhaps give engineers a $200/mo AI coding budget and let them pick from a selection of subscriptions. You’re going to have different engineers using different AI coding tools, and if you refuse to let them, your competitors will.
Maybe in the future “standardizing on one coding agent” will be a thing that makes sense. But that time is not now.
Are there security best practices for each of these tools or is that too early?
I don't have any use cases at all.
Business is definitely not booming.