Chris Hart

wish there was a /performance-review for claude code where i could get a token credit after it gaslights me for 80k tokens about the thing that's broken and how to fix it

@odd_joel Take my money! (Seriously, I paid for the lifetime plan yesterday.) Great job, can’t wait to see what you’re cooking next.

And this was before the watch app! 😻

@HighyieldHarry tired: mouse jiggler keeping teams status active wired: keyboard automation that launches excel, keyboard shortcuts to copilot, and asks to rebuild the same financial model it already built 300x

Got a scoop from a Wells Fargo employee that effective May 1, if you dont use CoPilot at least 3 days/week, your manager will receive an escalation email that you aren’t using AI/CoPilot. And so it begins.

@KevinNaughtonJr apps.apple.com/us/app/moshi-s… is worth a look. i had cobbled together tmux + cc + tailscale with moderate success, but adding the moshi ios app was a big unlock.

one of the many reason why you should do all your development remotely over ssh -open and close your devices and work continues in the cloud -connect and disconnect freely on all your different machines (including phone) and pick up exactly where you left off -increases your battery life since nothing runs locally -never need to reconfigure machines since the only one that matters is in the cloud -worry less about your stuff: if something gets lost/stolen/breaks you won't care (aside from cost) since you machines are just portals to your remote devbox -if you're ever without wifi you can work locally if you sync your files across devices with syncthing and changes are synced when you're back online -many other benefits of having a cloud machine like using tailscale with it as a VPN/exit node and much more try this workflow for a couple months and i guarantee you'll never go back to developing software locally

🚩 for success as an AI-native developer: having worked as a software developer and *really* liked the writing code bits holy trinity success predictor: previous professional software development experience + business acument + strong dislike for coding

claude code trying to prevent itself from making the same mistake in the future: I've added a one-line note on the function docstring so future-me remembers to keep them in sync. we're nearing AGI, it's just going to look like the guy from Memento

@ChrisLaubAI TLS fingerprint?

A Rust dev just killed Headless Chrome. It's called Obscura. The open-source headless browser purpose-built for AI agents and scrapers at scale. Chrome vs Obscura: - Memory: 200MB+ → 30MB - Binary: 300MB+ → 70MB - Page load: 500ms → 85ms - Startup: 2s → Instant - Anti-detect: None → Built-in Single binary. No Node, no Chrome, no dependencies. Stealth mode is brutal: → Per-session fingerprint randomization (GPU, canvas, audio, battery) → 3,520 tracker domains blocked by default → navigator.webdriver masked to match real Chrome → Native function masking so detectors can't sniff it out Drop-in replacement for Puppeteer and Playwright over CDP. Zero code changes. If you run agents or serious scraping at scale, this repo prints money. 100% Opensource.

gough : sf :: houston : nyc, i don’t know how else to explain it

Many entrepreneurs would be better off with a little more governance, particularly financial reporting. Most founders understandably resist this, because keeping control feels better, and there are many (true!) horror stories of founders getting thrown out of their own companies. But the practice and rigor of having to report performance, especially financial results, inoculates founders against some certain types of failures and errors in judgement. Governance and reporting in early stage companies gets a bad rap because a P&L isn’t the product. But it does help you stay alive long enough to create the product and make better decisions when it’s time to pivot, hang it up, or invest more. The discipline to pay attention to the numbers usually isn’t inherent in first time founders, which is why “governance” is a useful forcing function.

@TimHaines Everyone else saying training data is right, but the more helpful answer is that very clearly spelling out the tech stack with versions for a greenfield project helps get what you want. You can also remind it to check the docs for those versions if APIs or CLI changed.

Why is claude code giving me postgres 16, redis 7, and node 20 on a brand new project?

@GeorgeJourneys @mynamebedan Why can we assume there is no Chinese equivalent?

So, basically, if Anthropic was not a US company, we’d be facing zero days with multiple unknown points of attack on virtually all of our systems to an adversary who developed this capacity before us.

You could also use something like JFrog or a cloud providers private repos and mirror only trusted packages and package versions into it, then block access to public repos. Gives you a lot of speed, but forces deliberate evaluation of any newly introduced packages. (Also protects against supply chain attacks on existing, valid packages.)

Okay honestly this makes vibe coding into production very dangerous, you guys were all right I think what I'll do is cut off all access to DBs and run it as a user with almost no privileges

everyone talks about "proof of human", but would about "proof of agent" or "proof of token spend"? i can see some scenarios where a tool harness signing mcp interactions, or proving that tokens are being spent on an outcome, could be valuable

The quality of the Codex harness is superior and cloud vs local is more fluid. Claude Code has a much more robust skills ecosystem that is easy to add, and tool calling seems to work better for product planning. I wish all of the latter worked in a thoughtful GUI based harness that allowed seamless use of local and cloud agents. Only question is if OAI improves skills and planning tool calling faster than Anthropic improves their GUI harness.

90%-95% of AI discussions on X and LinkedIn are about Claude. No one shares skills or prompts for Codex. I don't see people helping with Codex setup. It's all Claude. That's the biggest vibe change in the past 3 months. And that's driving rapid revenue growth for them. Yes, their models are better in some areas, and the DoW stuff may have helped a little bit, but mostly Claude's approach of building vertical agents that give people "aha" moments quickly worked. Even if it's the same underlying model, just saying that you can access it differently - in the terminal or in an app, and messaging it differently for different verticals and use cases made it real for people. Most people don't care about far-fetched AGI claims. They want to use AI today for their use case. Claude Code captured the imagination of that group. And now they are just building on top - even if it's expensive. Even if ChatGPT 5.4 is as good as Opus 4.6 with 50% of the price. The product strategy and marketing approach made all the difference. It's their ChatGPT moment.

"Who thinks cost per token will keep decreasing?" was a question I asked the audience at a panel last year in SF. Almost everyone raised their hand. I was skeptical, and everyone looked at me like I was crazy. I think the Anthropic-$15-$25-per-PR pricing test is an indicator of what's to come. These AI tools are incredible, but I remain skeptical about the unit economics working at current prices.

any suggestions / best practices for using automations to loop over linear issues? triggering on issue create seems tricky because of dependencies on other issues which haven't been completed. maybe run on a schedule and include instructions to consider the dependencies between issues, then run a merge-conflict-resolution cycle after?

Cursor now has automations! You can run agents on schedules, trigged by events from Slack, GitHub, or any MCP server. I get a daily review every morning with my GitHub/Slack activity. Our team now has dozens of agents running 24/7 improving or monitoring things for us.

@badcop_ MapAnything was doing this years ago, Salesforce acquired them and integrated it in 2019: salesforce.com/news/stories/s…

free app idea for vibe coders!! salespeople tend to do a lot of travel. imagine they can put in all of the locations they need to visit for the year, and it plans out the shortest route that visits all of them once before finally returning home why has nobody built this yet?

@victorthigpen24 @dostoevesque not to sound like yoda, but… later, the enjoyment happens, first, the absence of expectation

@dostoevesque They asked for the secret to enjoying it and you basically said to not enjoy it so I don’t think that answers the question, because some people do seem to truly enjoy it

You have to stop treating it like a performance where you are trying to become the best version of yourself. It is not theatre. It is not self transformation day after day. You do it the way you brush your teeth. It is not about liking or disliking it. It is simply something you do. The moment you frame working out as a heroic arc, you introduce emotion into it. Motivation, inspiration, identity, self judgment. That makes it unstable. Some days you will feel like the protagonist. Most days you will not. But brushing your teeth requires no identity. No inspiration. No self analysis. It is maintenance. When exercise becomes maintenance rather than self-improvement theatre, it becomes lighter. Less ego. Less pressure. Less evaluation.

i guess it's partially preference, but the benefits i've seen: 1) it allows me to manage the work in linear, so if there are dependencies between features, those dependencies show up in linear and the skill avoids development on issues that are blocked 2) it decouples planning from development, so even if features are blocked due to dependencies on other work, the planning phase can still happen 3) it keeps claude code's plans for how to do the work in linear (where it can be reviewed and iterated on) without cluttering the source repo with a bunch of MD files that become unruly over time and add unnecessary content to the context if you wanted to manage all that in files in the source repo, or through issues in something like github with CLI integration for claude code, no need for the mcp server.

@ravikiran_16 @ceehart Why are MCP servers important to this workflow? I feel like Claude code can just do this with eg., git (+worktrees).

Tell me about your claude code or codex setups?

there are some practical limitations i've run into: 1) without some other agentic process for reviewing & merging PRs, the planning & dev agents run wild on the original main branch and lose coherence with the state the merged code will (eventually) be in, and i haven't gotten this part agentically-enabled 2) relatedly, but a smaller issue, is the same issue that comes up with regular engineers and competing, parallel database migrations would love ideas on solutions to these!

@ceehart @danveloper skills + mcp servers is where agents stop being autocomplete and become actual workflows. plan→implement→PR loop running without human shepherding is wild