robeym

Fun read! This is a good story to share. I spent over a year building a complex SaaS product for a narrow market, but it was something I personally needed and was interested in. It was a second full-time job as a solo developer, but the chances are if you are interested in something and find it worthwhile, others will too. It's worth it! Great work. Fun story. Thanks for sharing

Pull Requests, Issues, Git Operations and API Requests

I think quality experience and personal values are more important than engineer cost. I've seen so many shortcuts taken on outsourced work these past few years. AI also loves shortcuts. The combination of the two is not worth the cost savings.

If you value high quality work and pride in what you do, outsourced workers (who most often don't pay careful attention to their work, hence the cost) are not the solution. However, if you're just trying to get something done and don't care about it getting done right, what better way to do it than spending the least amount of money possible

I don’t think we should be discouraging people from buying homes. In a healthy market, the goal should be more households owning the homes they live in, not more homes being accumulated by investors and non-owner-occupant buyers.

The mistake is not buying a home. The mistake is buying too much home, stretching the term too far, and ignoring maintenance.

A cheaper house, shorter loan term, and realistic repair/addition budget change the math a lot.

Own what you can actually afford to maintain.

I was in the same boat, moving my max subscription to Codex instead of Claude, with a Claude Design project going. I was under the impression that Design was pro plans only, so I downloaded basically everything before cancelling.

It could be worth a quick $20 subscription just to grab your stuff, then cancelling. Trying to get support from either Claude or OpenAI seems pretty hopeless. Hopefully this post will get them to see you

I think progressing as humans is something to be proud of. I care less about who gets credit and more about what we can now do.

I also do not think this makes people less capable of solving hard problems. The bar just moves up. More people can now work on harder problems with better tools.

If the goal is credit or proving real skill, then focus on harder problems, like ones AI can't reach

I believe this is in response to PocketOS. When I read the original post, I was trying to figure out how they even built a workflow that had AI so close to the self-destruct button. This post's explanation about it probably being fully vibe-coded makes sense. How else would the system be so fragile and for the agent to have such far reach? They built a house of cards.

A far better approach is being precise with your prompts and if you find a new model has any bad habits, address it specifically in your AGENTS.md and go on your way. If you want to throw in slop promts, go ahead and add a massive AGENTS.md your employer gave you.

People waste too much time on this stuff. The next version could totally change how the model processes your agents.md.

Get good at promting, use agents.md as a minimal model annoyance fixer, and reset it often (every major release)

Anytime Microsoft buys something it's time to start looking for alternatives

Most people run to GitLab or Codeberg. The real loss is network effect of everyone being on GitHub.

GitLab: most similar to GitHub. Repos, CI/CD, issues. Hosted or self-host

Codeberg: free, non-profit, FOSS-friendly. Runs Forgejo. Where Zig went

Forgejo / Gitea: lightweight, self-host on a small server

Bitbucket: fine if you're already in Atlassian/Jira land

Sourcehut: minimalist, email-patch workflow. Fast but a culture shift

Radicle: peer-to-peer, no central server. Niche, principled

Tangled: newer, built on ATProto

Solid and growing impressions but low CTR. We're not quite on page 1 for most queries, and not even close for our primary target queries.

I've used Claude Code from the beginning and the first waves of changes were genuine improvements, with a very steady 4 months in late 2025. But recently, these past 2-3 months, the changes have shifted towards significant and frequently degradations in model performance. I had a very consistent workflow for 4+ months in Claude Code, but this year I've had way more surprises. I'm not sure which tools you use, but Claude Code had a great period of consistency, at least compared to other AI tools.

As a long-term 20x user, Claude has recently felt a lot like using AI for coding a year or so ago. It can't reliably handle basic tasks. I ask for something straightforward and get something subtly wrong, incomplete, or just not workable. I always use the best model available and effort levels maxed, but with all their changes I have to relearn how to make the model perform at best every day, and it seems I can't keep up. It’s not that Claude can’t do impressive things, it clearly can, but the inconsistency on simple, expected behavior makes it hard to use. The downtime is annoying but hasn't been the deciding factor. I’m not waiting it out this time. I’m switching over to Codex, and based on my usage today it looks like I’ll be fine on the 5x plan, so I can drop down and save about $100 a month which is nice. I didn't quite have a grasp on how quickly companies can change for better or worse until Anthropic showed me. I'm surprised at how quickly they brought me from a happily paying max user to not even wanting the lowest paid tiers.

Agreed. I wonder if large companies are already deciding on which tool to use based on guardrails. I'm seeing a lot of Copilot, but that's probably because of deep M365 roots they might've already had, rather than it being chosen for being reliably safe.

I don’t really get the logic behind letting agents run with full access to anything important. On one end you’ve got fully sandboxed setups where they can barely do anything useful, and where the user is scared to let it read files, and on the other end people are just pointing them at production systems and hoping for the best.

It's great to get excited about new tools, but learning how to use a tool before diving in is foundational.

I guess I'm asking if their CLI tool is the same or if it functions different. I've never used anything besides CC so I wouldn't know if it's basically the same thing

What's your workflow like? I'd be curious to test OpenAI out again but Claude Code is how I use the models. Does it require relearning another workflow?

I've always seen similarities between side project websites. It just depends on what the most popular/easiest to use tool is at the time. There was a huge surge of almost identical reusable components 8-10 years ago, but I didn't see nearly as many people talking about that as I do now with AI. People who have side projects will often go with what's easiest to rollout. It doesn't bother me at all. I don't understand why it bothers so many people so much .

Great job to everyone who has created something

I'm also noticing today that the model is hanging a lot. 5 min in, 50 tokens. Stuck in "Still here, still at it..."

Not being able to reliably control subagent model is the main reason I have it off.

Working on some research projects to test Opus 4.7.

The first thing I notice is that it never dives straight into research after the first prompt. It insists on asking follow-up questions. "I'd love to dive into researching this for you. Before I start..." The questions are usually silly, like, "What's your angle on this analysis?" It asks some form of this question as the first follow-up every time.

The second observation is "Adaptive thinking" replaces "Extended thinking" that I had with Opus 4.6. I turned Adaptive off, but I wish I had some confidence that the model is working as hard as possible (I don't want it to mysteriously limit its thinking capabilities based on what it assumes requires less thought. I'd rather control the thinking level. I liked extended thinking). I always ran research prompts with extended thinking enabled on Opus 4.6, and it gave me confidence that it was taking time to get the details right.

The third observation is it'll sit in a silent state of "Creating my research plan" for several minutes without starting to burn tokens. At first I thought this was because I had 2 tabs running a research prompt at the same time, but it later happened again when nothing else was running beside it. Perhaps this is due to high demand from several people trying to test the new model.

Overall, I feel a bit confused. It doesn't seem better than 4.6, and from a research standpoint it might be worse. It seems like it got several different "features" that I'm supposed to learn now.

Assuming /effort max still gets the best performance out of the model (meaning "ULTRATHINK" is still a step below /effort max, and equivalent to /effort high), here is what I landed on when trying to get Opus 4.7 to be at peak performance all the time in ~/.claude/settings.json:

  {
    "env": {
      "CLAUDE_CODE_EFFORT_LEVEL": "max",
      "CLAUDE_CODE_DISABLE_BACKGROUND_TASKS": "1"
    }
  }

The env field in settings.json persists across sessions without needing /effort max every time.

I don't like how unpredictable and low quality sub agents are, so I like to disable them entirely with disable_background_tasks.

Blogs -> trials/contact seem to be decent, depending on the post. The comparisons are new and inconclusive. LinkedIn posts are minimally tested, but some clicks have come in. Haven't tested X or Meta platforms yet.

I've handled all blog posts so far. Comparisons were passed to the team. They use various tools for research.

Identifying the long-tail keywords with clear buyer intent, and then ranking high for the query, is easier said than done. Blogs are my attempt to catch specific keywords that match what PAX offers. The results are trivial compared to ads, so maybe we still haven't realized the full potential of SEO + content angle.

I'm not interested in new features when the main ones are mysteriously getting worse and worse. I need to have a sense of stability before I get excited about any other features.

Thanks! The homepage has a short dashboard + payment video and the before/after section will show two basic dashboards as well. But I agree, we could use more static screenshots, expanding the /docs page with those is on the near-term list.

Feel free to reach out via the website forms to chat about ERP.

PAX ERP - full ERP + CRM for small manufacturers (5-50 people). Accounting, inventory, manufacturing, and CRM in one system. No QuickBooks dependency, no implementation fees, 3-day go-live.

Built it originally for internal use inside a medical device shop. The people running production were the same people responsible for financial accuracy, so every transaction (receiving a PO, issuing materials, completing production) auto-generates GL entries through database triggers. No configuration step, no separate accounting department required.

Current focus: SEO and content - competitor comparison pages (vs Fishbowl, vs Odoo, etc.), compliance landing pages for FDA/ISO 13485, blog posts targeting long-tail manufacturing ERP keywords. Building from 3 backlinks to 20-30.

Competing against Fishbowl (QuickBooks add-on), Odoo (powerful but a lot to configure), and the incumbent answer of spreadsheets + QuickBooks.

https://www.paxerp.com

So if it bypasses, is the optimal setting for performance setting effort level to max, keeping adaptive on? I try to avoid letting the model decide what is unimportant and needs less thought

Nothing super noticeable. I've reached 35% in sessions on the 20x plan. Before these changes, 25-30% was pretty normal. I think these changes are best for people who are just past the 5x usage plan, but might be harder to manage if you already have to throttle usage to stay under limits.

I'd still recommend turning off sub agents entirely because it doesn't seem you can control them with /effort and I always find the output to be better with agents off.

This is confusing. ULTRATHINK is a step below /effort max?

ULTRATHINK triggers high effort. /effort max is above high. Calling it ULTRATHINK sounds like it would be the highest mode. If someone has max set and types ULTRATHINK, they're lowering their effort for that turn.

For anyone reading this trying to fix the quality issues, here's what I landed on in ~/.claude/settings.json:

  {
    "env": {
      "CLAUDE_CODE_EFFORT_LEVEL": "max",
      "CLAUDE_CODE_DISABLE_BACKGROUND_TASKS": "1",
      "CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": "1"
    }
  }

The env field in settings.json persists across sessions without needing /effort max every time.

DISABLE_ADAPTIVE_THINKING is key. That's the system that decides "this looks easy, I'll think less" - and it's frequently wrong. Disabling it gives you a fixed high budget every turn instead of letting the model shortchange itself.

There are several settings in my account relating to Copilot that are locked/enabled with a shield and key icon next to it. Any idea how to disable these settings? It's on the same settings/copilot/features page.