Switch to opencode and get out of the vendor lock-in so you can use more models. Then try Deepseek v4 Pro and save massively without loosing AI ability. You won't run out of money and you'll get nearly the same capability. When the next great model drops you can easily switch to it regardless of vendor without switching tools
@dhh Have you experience K2.5 making more assumptions on its own than Claude Opus 4.6? I use it extensively on a .Net web application so it might have less training examples (I dunno) but I get frequent "additional features" from Kimi that I didn't ask for compared to Claude
@opencode how are the monthly subscriptions ($20/$100/$200) different from zen pay-as-you-go with auto-reload and monthly limit set? Are the tokens cheaper or are there other advantages to the monthly subscriptions?
If you are a dev in BigCo and reliant on BigCo to train you, you're not gonna make it. BigCo execs in charge of budgets, more often than not, know less about the current and future skill sets needed than a motivated dev does. If you're just looking for a paycheck then sure, just keep doing the assigned training in your HR portal, but if you love the craft and crave the knowledge then you better get plugged into the right communities and learn to teach yourself. Knowledge should never be paywalled and you should never let another person limit or control your desired future.
Don't fall for the "software engineering is dead" hype from the pro-AI crowd. It's software engineers who build AI, software engineers who create/orchestrate the ai agents to write software, and software engineers that are the human in the loop who review, correct, refine what the ai agents produce. It's like giving a calculator and/or spreadsheet to an accountant. It didn't eliminate accountants, it made them more efficient and accurate. After the AI hype is over and we begin to use it for it's true value, software engineers will be the puppet masters.
@jhansendk@mjovanovictech I don't have any real-world experience using it. Thank you for suggesting it as I'd love to see if/how it helps reduce ceremony. On the reading list
Same... entirely new set of skills to learn, pioneer un-proven workflows to see what helps vs. what doesn't (single agents or multi-agent, mcp or skills, huge https://t.co/OzPtCxU5BA or nested / linked docs, which model outputs the best for my given language/framework choice). Fun, but definitely more work, not less at this particular point in time.