Claude, how dare you delegate things to me when I am trying to delegate things to you
English
Derek Cheng
371 posts
Derek Cheng
@derekcheng
Founder @tonkotsudotai. ex-Atlassian, Microsoft, Meta.
Seattle, WA Katılım Ağustos 2010
1.1K Takip Edilen174 Takipçiler
Got laid off from Atlassian last week. Regardless of the speculation online, seems like a business decision more than anything else.
No factor. I've been building since that every evening.
Shipped accessibility-linter, a Claude Code skill that reviews a11y mid-design, not after the fact.
- catches a11y issues while you design, not after
- challenges the structure, not just the symptom
- covers ARIA, keyboard nav, focus, states, contrast, motion
- works with HTML and React
- runs as a conversation, not a report
github.com/designedbysom/…
Works with @claudeai Code, @cursor_ai , @Replit , @v0 , @figma Make, Rovo Dev. Give it a ⭐️ if you find it useful.
@designertom - I heard you. Sharing what I build on X. More interesting stuff coming soon.
English
"We're moving toward a world where the canvas and the code aren't separate things connected by a handoff"
This resonates. I think the next step is that generation and refinement both happen in the production codebase.
The power coding agents give to designers is to remove the disintermediation that resulted from having design artifacts that are independent of the code. When you're directly in the codebase, you can operate a lot more directly.
English
I think this is going to be a golden age of design:
1. The fundamental product design and development workflow is radically changing. Designers have way more agency and don't need to work through developers. Everything they do is more direct: design in the codebase, ship directly to customers.
2. Visibility bias. When all that's left is judgment and taste, the question is what should we spend our judgment and taste on? Visible/accessible elements of products will always get the largest share. Anyone who's been in an exec review will know the VP has ten points to make about the UI before asking a token question about the core infra.
English
What ratios are shifting as a result of coding agents?
- Coding : verification
- Coding : design
- Headcount, budget, influence of PM:eng, design:eng
- Customers/product footprint : team size
English
I haven't seen much written about this, but I'm guessing coding agents are driving an increase in monorepos. Rather than specifying an API and having Claude implement it in a repo, then consume it in another, just let it figure out the end to end.
English
Huge next round of innovation: when coding speeds up 10x, what becomes the new bottlenecks?
Nikita Bier@nikitabier
iOS developers: How long is App Review taking for everyone these days? It is now taking longer to get our app approved than it is to build the actual features.
English
This resonates 100%.
Coding agents have created the opportunity for designers to assert a lot more agency: no more being one-layer-away from the production product and customers.
Also, the "taste and judgment" role that humans play is going to always be more tilted toward visible elements. Yeah, we'll supervise the backend architecture as well, but not nearly to the same degree of detail.
Both the above create huge opportunities for designers.
English
@auchenberg Yep, and plan reviews are the new code reviews. Such a high leverage point to influence things.
English
The best thing the public faces of Claude Code could do right now as this blows up - is come out with a simple - developer and human friendly - explainer on what is and is not allowed. No legal mumbo jumbo - just be up front and clear with your customers. List the 20 common scenarios you see in the wild and just yay or nay them.
Rob Zolkos@robzolkos
Major Claude Code policy clear up from Anthropic: "Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted"
English
@GergelyOrosz Yeah, strange move if they want to own the agent platform layer (which I thought they did).
English
Huh.
So Anthropic doesn’t want an ecosystem built on top of their SDK.
Message received.
Between them banning OpenCode, their hostile stance for OpenClaw, and now this: Anthropic make it they don’t want 3rd parties to build on them.
Bug win for eg Codex, MinMax and others…
Rob Zolkos@robzolkos
Major Claude Code policy clear up from Anthropic: "Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted"
English
@esthercrawford There's so much to be built with the goal of scaling that upper bound for simultaneous work. That feels like the primary unlock for the next little while.
Our attempt in the coding space: @tonkotsudotai
English
Working on 1 project at a time is painfully slow due to downtime in between the prompt & code output so I need multiple projects going at once to feel productive. I wonder, what's the upper bound for simultaneous work given the cognitive load of context switching as a human?
English
People have stopped talking about p(doom) lately.... even though AI is more capable than ever now.
What happened?
English
Someone's gotta be working on a visual 'IDE' geared at:
- collaboratively writing tasks & requirements
- structured queuing of work
- understanding & navigating code w/vector indexes
- navigating, reviewing & commenting on changes
No code editing required.
Anyone?
English
Couldn't agree with this more. Another common mistake/lesson for new managers: calibrating when to let go and when not to. Let go too little and you micromanage and can't scale. Let go too much and things go to hell.
There will be so much impact over the next few years from teaching ICs to manage for the first time.
English
The explanation for this is quite simple. Many ICs just became managers with zero experience or training. Management itself is a skill to learn. An extremely common mistake new managers make is to confuse "feeling busy" with "impact".
Rohan Paul@rohanpaul_ai
A super interesting new study from Harvard Business Review. A 8-month field study at a US tech company with about 200 employees found that AI use did not shrink work, it intensified it, and made employees busier. Task expansion happened because AI filled in gaps in knowledge, so people started doing work that used to belong to other roles or would have been outsourced or deferred. That shift created extra coordination and review work for specialists, including fixing AI-assisted drafts and coaching colleagues whose work was only partly correct or complete. Boundaries blurred because starting became as easy as writing a prompt, so work slipped into lunch, meetings, and the minutes right before stepping away. Multitasking rose because people ran multiple AI threads at once and kept checking outputs, which increased attention switching and mental load. Over time, this faster rhythm raised expectations for speed through what became visible and normal, even without explicit pressure from managers.
English
@kunchenxyz I think the thing that remains to be proven out is if you can build a thick enough product/business as an extension (i.e. does my #2 exist?)
English
@derekcheng Yeah it seems the overwhelming trend is that people bring various extensions into these “master agents”. And the extensions are largely free commodity.
Independent “agents” that are getting traction are usually because of UX.
English
The saaspocalypse takes that "I can rebuild salesforce in a weekend" miss the real threat. It's not replacement, it's disintermediation.
It looks quite likely that the core interaction model with AI will be agents. The frontier labs are assembling agent platforms and will work to aggregate a critical mass of agents.
When that happens, the primary mode of interacting with a saas will be via an agent sitting on someone else's platform - classic disintermediation, risking the saas incumbent being reduced to a data store or workflow engine backend.
In this world, it seems there are three new opportunities:
1. Be the agent platform that dominates by aggregating the most agents
2. Be a top agent builder on said platform
3. Build a UX layer to interact with all these agents
English
@kunchenxyz It's true that it's early. Skills and MCPs are a bit thin. Agent SDK allows for a lot more but you're right that we haven't seen massive killer apps on that yet. I assume there will be extension points in Cowork as well.
English
@derekcheng A really interesting factor is that it’s not clear “agent builder” is actually a thing yet. Outside of the LLM and UX, there really isn’t much differentiation outside of some niche segments where private domain data is critical (harvey etc).
English
Man, I'm tolerant of hacks, but Sonnet buddy, we gotta talk about this one.
English