The “typing” part used to dominate the cost structure, so we optimized around it (architecture upfront, DRY everywhere, extreme caution). Now the expensive part is clarity of intent and orchestrating the iteration: deciding what to build next, what to cut, what to validate, what to trust, and where to add guardrails (tests, invariants, observability).
If your requirements are fuzzy, the agent will happily generate 5k lines of very confident nonsense. If your domain model + constraints are crisp, results can be shockingly good.
So the scarce skill isn’t “can you write good code?”
It’s “can you interrogate reality well enough to produce a precise model—and then continuously steer the agent against that model?”
What works extremely well for me is this: Let Claude Code create the plan, then turn over the plan to Codex for review, and give the response back to Claude Code. Codex is exceptionally good at doing high level reviews and keeping an eye on the details. It will find very suble errors and omissins. And CC is very good at quickly converting the plan into code.
This back and forth between the two agents with me steering the conversation elevates Claude Code into next level.
I wonder whether we will see a shift back toward human generated, organic content, writing that is not perfectly polished or exhaustively articulated. For an LLM, it is effortless to smooth every edge and fully flesh out every thought. For humans, it is not.
After two years of reading increasing amounts of LLM generated text, I find myself appreciating something different: concise, slightly rough writing that is not optimized to perfection, but clearly written by another human being
If LLMs presently aren't capable of matching the style quirks you're describing, isn't it likely they'll be able to in the near future? To me this feels like a problem that'll either need to be addressed legally or left to authors to somehow convince their audiences to trust that their work is their own.
Everyone keeps arguing that AI is not Apple’s core business and that their priorities are different. From an end-user perspective, that is irrelevant.
What users actually experience is this: every other major platform is shipping increasingly capable intelligent assistants. These systems can interpret intent, execute multi-step actions, and meaningfully reduce friction. Meanwhile, Siri still struggles with fairly basic workflows.
At the end of the day, I do not particularly care about internal constraints, organizational structure, privacy positioning, or strategic rationale. What matters is whether the product works.
Today, I still cannot reliably:
- Dictate complex voice input without constant correction
- Use voice to control my iPhone in a composable way such as “open this contact and send a message,” “replay the song I liked yesterday,” or “create a note in Obsidian with this content: …”
- Chain actions together in a way that reflects actual user intent
These are not futuristic requests. They are practical, everyday workflows that competitors are increasingly able to handle.
The gap is no longer about incremental feature parity. It is about whether Apple can deliver a genuinely intelligent interface layer, or whether Siri remains a deterministic command parser in an era where users expect contextual reasoning.
But you can already do that, in the terminal. Open your favourite terminal, use splits or tmux and spin up as many claude code or codex instances as you want. In parallel. I do it constantly. For all kinds of tasks, not only coding.
Many of us have an aging neighbor whose hearing gradually worsens. The TV volume creeps up over time.
A simple, thoughtful fix is to gift them a wireless TV speaker designed for this exact problem.
The Sony SRS-LSR200 sits close to the listener, so dialogue is clear without blasting the TV for everyone else. It lets them enjoy their shows again without turning the volume knob into a neighborhood event.
And of course they will demand that everyone is required to do a KYC. At sone point vpn access will require that as well. And finally the internet as we know it will be a thing of the past.
Starting in 2025, the European Commission is implementing a "mini-wallet" app (an Age Verification App Blueprint) designed to allow users to prove they are over 18 without disclosing their identity or exact age.
The US has already employed its technology and financial instruments, including sanctions, to coerce and control its partners. Sanctioning an ICC prosecutor and subsequently restricting Microsoft’s access to his emails and documents are just a few instances of this. They have demonstrated their willingness to use their technology, financial instruments, and sanctions against their partners. It seems almost too late for Europe to achieve its independence in both technological and financial spheres.
OT: Has anyone observed that Claude Code in CLI works more reliably than the web or desktop apps?
I can run very long, stable sessions via Claude Code, but the desktop app regularly throws errors or simply stops the conversation. A few weeks ago, Anthropic introduced conversation compaction in the Claude web app. That change was very welcome, but it no longer seems to work reliably. Conversations now often stop progressing. Sometimes I get a red error message, sometimes nothing at all. The prompt just cannot be submitted anymore.
I am an early Claude user and subscribed to the Max plan when it launched. I like their models and overall direction, but reliability has clearly degraded in recent weeks.
Another observation: ChatGPT Pro tends to give much more senior and balanced responses when evaluating non-technical situations. Claude, in comparison, sometimes produces suggestions that feel irrational or emotionally driven. At this point, I mostly use Claude for coding tasks, but not for project or decision-related work, where the responses often lack sufficient depth.
Lastly, I really like Claude’s output formatting. The Markdown is consistently clean and well structured, and better than any competitor I have used. I strongly dislike ChatGPT’s formatting and often feed its responses into Claude Haiku just to reformat them into proper Markdown.
Curious whether others are seeing the same behavior.
Do people actually browse the App Store to discover what’s new? I personally only open it when I already know exactly what I want to download, for example Obsidian or Firefox. I search, install, and I am done. I never scroll around or browse for inspiration.
I am genuinely curious how others use it. Is App Store browsing a real behavior, or is discovery mostly being forced because search no longer reliably gets you to the thing you already know you want?
The real bottleneck isn’t writing (or even reviewing) code anymore. It’s:
1. extracting knowledge from domain experts
2. building a coherent mental model of the domain
3. making product decisions under ambiguity / tradeoffs
4. turning that into clear, testable requirements and steering the loop as reality pushes back
The workflow is shifting to:
Understand domain => Draft PRD/spec (LLM helps) => Prompt agent to implement => Evaluate against intent + constraints => Refine (requirements + tests + code) => Repeat
The “typing” part used to dominate the cost structure, so we optimized around it (architecture upfront, DRY everywhere, extreme caution). Now the expensive part is clarity of intent and orchestrating the iteration: deciding what to build next, what to cut, what to validate, what to trust, and where to add guardrails (tests, invariants, observability).
If your requirements are fuzzy, the agent will happily generate 5k lines of very confident nonsense. If your domain model + constraints are crisp, results can be shockingly good.
So the scarce skill isn’t “can you write good code?” It’s “can you interrogate reality well enough to produce a precise model—and then continuously steer the agent against that model?”
reply