Creator of Claude Code Reveals His Workflow 54
Boris Cherny, the creator of Claude Code at Anthropic, revealed a deceptively simple workflow that uses parallel AI agents, verification loops, and shared memory to let one developer operate with the output of an entire engineering team. "I run 5 Claudes in parallel in my terminal," Cherny wrote. "I number my tabs 1-5, and use system notifications to know when a Claude needs input." He also runs "5-10 Claudes on claude.ai" in his browser, using a "teleport" command to hand off work between the web and his local machine. This validates the "do more with less" strategy Anthropic's President Daniela Amodei recently pitched during an interview with CNBC. VentureBeat reports: For the past week, the engineering community has been dissecting a thread on X from Boris Cherny, the creator and head of Claude Code at Anthropic. What began as a casual sharing of his personal terminal setup has spiraled into a viral manifesto on the future of software development, with industry insiders calling it a watershed moment for the startup.
"If you're not reading the Claude Code best practices straight from its creator, you're behind as a programmer," wrote Jeff Tang, a prominent voice in the developer community. Kyle McNease, another industry observer, went further, declaring that with Cherny's "game-changing updates," Anthropic is "on fire," potentially facing "their ChatGPT moment."
The excitement stems from a paradox: Cherny's workflow is surprisingly simple, yet it allows a single human to operate with the output capacity of a small engineering department. As one user noted on X after implementing Cherny's setup, the experience "feels more like Starcraft" than traditional coding -- a shift from typing syntax to commanding autonomous units.
"If you're not reading the Claude Code best practices straight from its creator, you're behind as a programmer," wrote Jeff Tang, a prominent voice in the developer community. Kyle McNease, another industry observer, went further, declaring that with Cherny's "game-changing updates," Anthropic is "on fire," potentially facing "their ChatGPT moment."
The excitement stems from a paradox: Cherny's workflow is surprisingly simple, yet it allows a single human to operate with the output capacity of a small engineering department. As one user noted on X after implementing Cherny's setup, the experience "feels more like Starcraft" than traditional coding -- a shift from typing syntax to commanding autonomous units.
with less? (Score:5, Insightful)
1 Boris, 5 Claudes in terminal, 10 Claudes in Cloud
I wouldn't call the datacenter required to run those Claudes in the Cloud 'less' than say 16 laptops that 16 devs would use.
Re: (Score:1)
I wouldn't call the datacenter required to run those Claudes in the Cloud 'less' than say 16 laptops that 16 devs would use.
Except you're not paying for 16 devs nor using 16 laptops. You're using one person on one machine acting as multiple people and machines.
Re: (Score:2)
That requires a megawatt of power, sure efficiency, right.
Do you know the difference between power and energy? Necessarily, even if it spends the same amount of energy, if it gets done in less time, it will use more power. That's how power is defined, it's the derivative of energy. So, that metric is already useless.
Now, if you also understand that getting things done faster has a value additional to getting the same thing done in more time, then now you've justified spending more energy as well.
The track for handling global warming needs to be less humans, not le
So, let me see... (Score:5, Insightful)
This is a commercial by a peddler of some software about how good that software is, supported by two unknown shills introduced as "a prominent voice" and "an industry observer"?
Ooookay, slashvertisement department, you could not be any more transparent.
Re: (Score:2)
Well, it is AI slop from venture beat, what did you expect?
Re: (Score:2)
I agree with you. This article is saying exactly what employers want to hear, in a very matter-of-fact and highly enthusiastic way. "Oh sure, one guy does the work of 10 now. It's totally easy and anyone can do it. Just buy my product!" It's not, however, providing any kind of objective metrics that this is even true, nor indicator that this productivity level applies to all project types, nor whether the code stays maintainable in the long term.
The article states that opting for this pricier bot and p
Re: (Score:2)
Those of us who don't work for anthropic, and whose bean counters are large and in charge will definitely not appreciate 5 (or more) free-running claudes. Either you'll spend a lot of money, or they'll just impose a token limit and 5 claudes will just be 5 hung terminals. If it's important, you'll have to go through a corporate approval/procurement process to have the cost approved.
I'd rather just write code than spend all day sell-sell-selling.
TFA sounds like Co-pilot orchaestration (Score:4, Informative)
TFA sounds like simple orchestration compared to Co-pilot [github.blog]. At least co-pilot allows your 'team' [visualstudio.com] to have various skills [microsoft.com] as they work cooperatively with each other. In other words, the technology is similar yet co-pilot has a whole API developed.
Re: (Score:2)
...a developer agent might use Anthropic Claude, a documentation agent might use Gemini, and a testing agent use Codex, because those LLMs have their specialties that set them apart. Co-pilot manages the team.
Re:TFA sounds like Co-pilot orchaestration (Score:4, Interesting)
Claude is very good at documentation and general text generation. I frequently have Claude write documentation in MD files describing the plan, detailed API docs, progress reports, etc. For the most part it's coherent and not as slop-sounding as I expected. Sometimes I have it write git commit messages which I have to say are often way better than the ones I write. More descriptive yet more concise than my own style. I just have to remind it to not use emojis in the messages (no icons basically). Claude loves to use emoji icons such as check marks.
Re: (Score:3)
I'm curious if you've also tried other LLMs for writing API docs, maybe Gemini (which in general seems very good), and how Claude compares specifically for that (writing detail-orientated documentation)?
Are you talking about Claude (web interface) or Claude Code (which works a bit differently in the way it edits files)?
Re: (Score:2)
Have not used Gemini so I can't compare to that. I'm sure they are somewhat comparable.
I primarily use Claude Code cli. I do not use Claude Code web.
In the Pipe, Five by Five (Score:5, Funny)
"feels more like Starcraft" than traditional coding
YOU MUST CONSTRUCT ADDITIONAL POINTERS
Re: In the Pipe, Five by Five (Score:5, Funny)
Re: In the Pipe, Five by Five (Score:1)
Re: (Score:2)
Yeah, I'm definitely aware of the original reference. I always like the crossover, but also wondered if it was legal. Looking back, it's seems trivial.
The only problem, though, is getting to that point (Score:5, Insightful)
See... for that to actually work for you... you already need to be an expert... otherwise you're going to vibecode yourself into a national security incident.
Re: (Score:3)
Re: (Score:1)
Re:The only problem, though, is getting to that po (Score:4, Informative)
Cherny does say on Twitter than he ends up throwing away 10-20% of the Claude Code sessions that he starts because they have ended up nowhere (he's not specific about the various ways it fails). No doubt as creator of the tool he's well attuned to knowing when to have it try to course correct and fix, or better to abort.
Baloney (Score:2)
I've tried it with 2 and it chokes.
Re: (Score:3)
You don't have unlimited access to an entire data center. He does.
Re: (Score:2)
What do you mean by "chokes"? You hit your usage limit quickly?
dear god (Score:2)
Re: (Score:2)
Claude has been generating pretty good Qt code for me. Most of it is boiler plate of course. I haven't personally had it litter my code with exception handling, but I have seen it do that to others.
You can always add to your CLAUDE.md file a note to tell it to avoid using exception handling. I'm pretty sure you could tell it to only catch exceptions that the code specifically is set to deal with, and let all others go on up.
Re: dear god (Score:3)
Or you could tell it to behave like someone who actually can code, and check all pre- and post-conditions, and invariants, so you don't get exceptions.
Re: (Score:2)
What tool do you suggest to generate code, that doesn't have to be managed?
That's nice (Score:2)
the creator of Claude Code ...
"I run 5 Claudes in parallel in my terminal,"
How did he get to the first Claude? And is that all it's good for? Creating more Claude? What else can it do?
Could Cherny maybe hop over and fix some of the bugs in systemd?
Re: (Score:3)
Re:That's nice (Score:5, Funny)
Claudes all the way down.
Like Slashdot history, clods all the way down.
Some of them insensitive.
Re: (Score:2)
Mod parent funnier.
expensive (Score:4, Interesting)
Anthropic has a bunch of legacy models you can use, but the recent Claude Opus 4.5 is the best coding assistant I've tried by far. Better than Google's Gemini, better than OpenAI GPT-5.2.
Unfortunately it is also relatively expensive, 2-4x the price of much of the competition. This guy runs 5 of them to help him and I bet they really do. I use one to do planning now and then or when the lesser models get stuck. I could burn through a several $100/month with 5 Claudes but it would be worth it in some situations.
Re: Doing more with more (Score:3)
Humans also consume a lot of resources. If we need fewer bodies over time, things may balance. The problem is the transition. Do we plug every redundant person into the Matrix ?
Re: Doing more with more (Score:2)
the number of humans is not a function of the "need" for them.
Re: (Score:2)
Supports, not validates (Score:2)
Re: (Score:2)
> I doubt it is really possible to produce quality output wrangling 5 sessions on 5 pieces of system code at a time.
Right, you can run as many Claude Codes in parallel as you like (each working on a different git checkout), but it's the human spinning them up and directing them, checking/testing the output, throwing away if necessary (Cherny says he throws away 10-20%), and a human swivel-chairing between 5 different development tasks at the same time doesn't sound like a recipe for high quality output.
Boris Cherny doesn't pay for tokens (Score:5, Insightful)
Try this yourself and see how fast it'll burn through your subscription / rack up API bills.
Re: (Score:2)
"I use Opus 4.5 with thinking for everything" (https://x.com/bcherny/status/2007179838864666847?s=20) wow that's direct evidence that he has unrestricted billing. Opus 4,5 will definitely shut down your allocation in a hurry.
Re: (Score:3)
I had a single Opus 4.5 just working on a simple script I didn't feel like writing. Burned the whole subscription in 2 hours. I only asked it 10-12 prompts.
more is not always better (Score:3, Interesting)
I also had occasions, when I had to tell claude, that something was already solved at some other place in the code, and it should reuse that. Sometimes, I have to tell this multiple times before the AI accepts that there is no new code needed.
Still, I find the tool really handy to get into a subject quickly. The main work changes to doing a) Designs, b) Reviews, c) Testing. Unlike other people, I always found the code easily readable and well documented. Therefore, after having done the Design, and knowing what should happen, the review is not too hard. Testing is a different story, but testing should take half of the time a project needs anyways (according to the mythical man month, which I consider still valid).
don't share X / Twitter links (Score:4, Insightful)
Slashdot: please stop sharing links to the sites that disseminate X links. Does Boris use Mastodon? X CEO Elmo Musk is a little fascist and a Trump ally who really wants to paywall the entire Internet through his own profit motives. To hell with Musk.