turnsout

As someone who would be in a position to advise enterprises on whether to adopt Frontier, there is simply not enough information for me to follow the "Contact Sales" CTA.

We need technical details, example workflows, case studies, social proof and documentation. Especially when it's so trivial to roll your own agent.

100%, particularly B2B SaaS

People really seem to believe that code is the only thing you need to make a SaaS company. It's like thinking a line cook is all you need to open a restaurant. There are so, so many other components to running a business.

“Human directing an agent” will become the dominant paradigm. We’ll still be in the loop, but there is no need for me to go to five different websites to look up basic information and synthesize the answer a simple question.

I wouldn't say it was oneshotted, but it did produce a working MVP in one Plan execution. Meaning, I went back & forth a few times about requirements, it built a plan, and then CC spent just under 15 minutes writing the code. Once I got the credentials plugged in, the core integrations (Slack, gmail, IMAP, iCloud calendar) and agent loop did work. I can share the initial message if you're curious.

Oh interesting—how do you find OpenCode vs CC? I'll check it out. And I'll try to get a version of this assistant in a form I could share publicly.

This is really awesome. I want to give my agent access to basic coding tools to do text manipulation, add up numbers, etc, but I want to keep a tight lid on it. This seems like a great way to add that functionality!

A lot of the system prompt, skills and tools center around my specific needs (I manage separate IMAP and Gmail inboxes, use Granola, and have iCloud calendars). And there are some hard assumptions baked in (I want to have a morning & afternoon check-in). It probably wouldn't be useful as-is, but maybe as inspiration?

Yeah, I looked at Clawdbot / OpenClaw at the beginning of the week (Monday), but the token use scared me off.

But I was inspired to use Claude Code to create my own personal assistant. It was shocking to see CC bang out an MVP in one Plan execution. I've been iterating it all week, but I've had it be careful with token usage. It defaults to Haiku (more than enough for things like email categorization), properly uses prompt caching, and has a focused set of tools to avoid bloating the context window. The cost is under $1 per check-in, which I'm okay with.

Now I get a morning and afternoon check-in about outstanding items, and my Inbox is clear. I can see this changing my relationship to email completely.

Benchmarks measure what they measure. But your subjective experience also matters.

Good-faith answer: I can't be certain. But I've been using CC since its release, and Cursor before that (and actually going all the way back to GPT3 to do codegen in the Playground). After getting used to the CC workflow, the way that I use it has been pretty consistent. To be specific, I use basically the same AGENTS.md with small modifications for each project, and I live almost exclusively in Plan mode and the best model (currently Opus 4.5).

My initial prompting is boilerplate at this point, and looks like this:

(Explain overall objective / problem without jumping to a solution)

(Provide all the detail / file references / past work I can think of)

(Ask it "what questions do you have for me before we build a plan?")

And then go back and forth until we have a plan.

Compared to my work with CC six months ago, it's just much more capable, able to solve more nuanced bugs, and less likely to generate spaghetti code.

This is probably entirely down to subtle changes to CC prompts/tools.

I've been using CC more or less 8 hrs/day for the past 2 weeks, and if anything it feels like CC is getting better and better at actual tasks.

Edit: Before you downvote, can you explain how the model could degrade WITHOUT changes to the prompts? Is your hypothesis that Opus 4.5, a huge static model, is somehow changing? Master system prompt changing? Safety filters changing?

It seems to depend on FlashAttention, so the short answer is no. Hopefully someone does the work of porting the inference code over!

It's always valuable to have a generalizable skill. But design is fundamentally a craft; an applied art. It's problem-solving. And like any craft, there are tools and techniques that are tried and true. You could approach woodworking with a ground-up Systems Thinking approach, but would you turn down the advice of a carpenter with 30 years of experience? Technically all you need to understand woodworking is a physics textbook and maybe an organic chemistry textbook.

My guess is you're a software developer (as I am), and in my opinion the fatal flaw of our group is the incorrect belief that we could do anything or solve any problem by simply decomposing it into smaller and smaller components. The thing is, for a big enough problem, there are an almost infinite number of ways to break it down and then build it back up. In optimization terms, complex projects are highly nonlinear problems, so you may be able to understand what the inputs are, but it sometimes takes wisdom and experience to tune the parameters.

I practiced Design Thinking at IDEO for 10 years, and I can assure you it's not "one size fits all." And you can onboard an intern or a client CEO in days, without requiring them to internalize a very abstract system for decomposing problems.

Foundation model sponsors already pay humans to generate authentic content, especially in technical areas that are underrepresented in general internet scrapes. I would imagine that this trend will continue.

Further, the "model collapse" hypothesis of 2020/2021 seems to have failed to materialize. Maybe we're still too early, and we're not yet seeing negative effects of OpenAI training on OpenAI output. But maybe "slop" is not being rewarded as much as human content, and having humans in the loop (even as readers) is preventing a slide into incoherence.

Will LLMs eventually disincentivize people from producing and publishing new original content? If that content is easily replicated by an LLM query, maybe. And maybe it's not the worst thing in the world. 5 years ago I would have bought an "FFmpeg Cookbook" from O'Reilly, but now I would just tell Claude exactly what I'm trying to achieve. As a consumer, I'm better off, and arguably we've saved the author of a hypothetical FFmpeg Cookbook weeks out of their precious life. Weeks they could spend doing something—anything—more valuable than rewording FFmpeg documentation.

Design Thinking (and more broadly, human-centered design) is a pragmatic framework for doing product design in an effective and productive manner. Systems Thinking is a massively more general superset. I'm not really sure how you'd operationalize that on a design project, except by following first principles, which would essentially get you to DT / HCD.

I strongly believe that users will start to demand proof of humanity (and related things like content credentials on photos/videos), and eventually filter out anything that isn't verified. Through great expense and loss of privacy, we will eventually claw our way back to an internet that looks more or less the same as it was in 2020.

Unfortunately, less privileged users will have to endure the sea of AI content that still preys on the unauthenticated. It will be like using the web without an ad blocker, but 1000X worse.

"Commingling" is such a great euphemism for fraudulent counterfeiting.

I can't count the number of times I've ordered a book from Amazon (1st party, Amazon as the seller) and received an obvious counterfeit, with fuzzy text and a poorly printed cover. On one occasion, the scanning/OCR process had missed most of one chapter, so there were just section headers, page numbers and blank pages.

Unfortunately publishers and manufacturers don't have a lot of leverage with Amazon. If there's pressure coming from somewhere, it must be coming from a regulatory body.

Yes, in Amazon's world, the star rating is for the product. Which is especially confusing on product pages with vastly different variants.

This seems like a predictable pendulum swing. I love AI, but I also love the phenomenon of people turning more toward IRL and tangible activities.

Anecdotally, I have friends who have recently bought turntables out of the blue and gotten into vinyl. Other friends who never had any interest in my analog cameras are asking about film. My wife has even switched from scrolling Instagram at night to working on a crossword book with a pencil.

None of them have put it exactly this way, but in divisive times, I think social media is just exhausting. And now you can't even really tell what's real.

In related news, I bought a CD player today

I'm assuming this is in the context of an agent that can see what it's doing. And I wouldn't assume humans have a monopoly on judging leading.

The Claude Code model highlights the power of simple search (grep) and selective reads (only reading in excerpts). The only time I vectorize is when I explicitly want to similarity-based searching, but that's actually pretty rare.

> Vocabulary is just the surface.

Yes, but with LLMs, sometimes simply mentioning the right words is enough to prime the model in the direction you want to take it. If you start a prompt talking about leading and type pairings, it will take greater care with typography. You don't need to be an expert typographer to take advantage of this phenomenon.

Honestly it sounds like they went above and beyond. Does this solve the trifecta, or is the network still exposed via connectors?

Honestly lack of true multithreading (without the Web Worker hack) is the biggest downside for me. Every major project I work on needs the concept of a main thread for UI and a separate thread for processing.

Looks interesting, but operator overloading is an anti-pattern. Leading with that is like leading with "full support for null pointers."

Now just get Claude Code to build a hUGETracker exporter, and you could actually bring one of these songs into GB Studio!

Just a hot take, but if you ask someone to complete a rote task that AI can do, you should not be surprised when they use AI to do it.

The author does not mention whether the generated project plan actually looked good or plausible. If it is, where is the harm? Just that the manager had their feelings hurt?