Michael Timbs

@zenorocha @resend This is a smart move IMO. I built @bourd_dev this way from day 1. Today there are even a few things that are *only* available via MCP and API and not via the UI. The only thing I deliberately excluded from this is minting new API keys.

we're starting a new project internally... it's called *The Headless Dashboard Initiative* the goal is to make sure that everything you can do on the @resend dashboard, you can also do via: • API • SDK • MCP question: which endpoints should we build first?

@ibuildthecloud I have no doubt that Pi is a much better harness. My concern is that I will not pay API pricing for frontier models and I don't trust that open source models are good enough yet.

I have serious fomo using something like pi/opencode. Like, "what if claude code/codex built in prompts are just better." How will I ever know. Or they have better tools. Like claude-code `/verify` is a hidden gem. Do you all not worry about this.

Imagine being a divorce lawyer and you get hit with 1000 openclaw instances bombarding you with emails.

@ThePrimeagen with a bit of work I think you could get that to 80%

Guys... I take back everything I have ever said. I am a terrible programmer. My game is spending 63% of its time in mouse entries map setting / getting...

I think frontier AI lab funding should only come through revenue generated from a reality tv show like battlebots where the labs have to have their AI manufacture and control robots that fight it out to the death on tv. An AI gladiator if you will. They could build a great colosseum in SF and hand out free fentanyl to the public on the days of the games.

@jiratickets Also password sharing for his Netflix

he should try making coffee at home

@rvcas Odin going to be big in 2027. I can feel it

I use typescript and rust everyday but I’m very Odin curious atm

@sama I interviewed at this startup about 6-7 years ago in fact

this would have been a whole startup not too long ago

@coffman_r In my experience junior engineers almost always ask for review on most changes even when it’s not mandatory. The frequency decreases as their confidence improves.

@michael_timbs Do you not work with junior engineers? They need their code reviewed, it’s a great opportunity to mentor them and help them grow.

This is a sign of a low trust team. If people are voting no here there are structural issues that need to be resolved. 75% of the team doesn't trust the code quality and decision making of their peers

@nayshins Before work from home Friday afternoons were reserved for sobering up enough to be able to drive home

I dont know who needs to hear this, but you really should never schedule an interview on a friday afternon

@itsmesatwik_ @DeepknowledgeU lol. Rightfully so

@michael_timbs @DeepknowledgeU You don’t fail the audit but a competent auditor will make a note if your change management is a stamp from Claude lol

How to set yourself up for failure for a SOC2 audit (and thus lose your license to sell to enteprise)

The more I use AI to try and brainstorm or do creative work the more I realise it’s not very good at this. It really can’t escape a fairly narrow set of concepts. “Memory” makes this way worse.

Me at 20: lots of ideas, not much experience Me at 35: no ideas, lots of experience

He talks about his wife like this too.

@robdel12 I've been through SOC2 four times as well as ISO 27001. You can absolutely 100% be compliant without mandatory human review x.com/michael_timbs/…

@michael_timbs None of yall work on anything that requires compliance and it shows. Or just openly violating SOC2

@lewiscarhart this is always the (uninformed) default argument from people who don't want to engage

if your soc 2 or iso 27001 auditor freaks out about this they are a boomer decel get in touch with the comp ai team and we'll make sure you're good to go 🤝🤝

@spirodonfl Rust + SolidJS for me but React is fine

Honestly, if you claim to care about code as a craft but you use anything except for React and Rust (driven by AI) then you're lying. You are *literally lying* There is zero percent chance you believe you care about code if you're going to ignore a programming language that (a) has famously been used to rewrite mass portions of Linux (b) has famously been used to rewrite Bun and (c) guarantees 100% memory safety with no effort at all, guaranteed, 100% Being a furry is not required. People need to stop harping on this. Furries make up only a large large large LARGE vocal portion of Rust but not all of them are like that. If you think you're supporting some social/political ideology by using Rust then you're out of touch. Besides, who cares who pushes for Rust these days? The language is from God. Just use it. If you're still doing "document.getElementById" then you don't care *at all* about code when you can have React manage all your DOM states for you with 100% perfect efficiency, guaranteed, with no effort at all. Do you think you're better than Facebook / Meta? Really? Those engineers make *bank*. Have you seen their paychecks? What's *your* paycheck like? Paycheck to paycheck are you actually going to tell me you're better than them? I don't understand how this industry can lie so much about everything.

I guess my point is you have to draw a line somewhere - e.g why not require that every engineer in the company reviews every PR. My argument is that the person who carries out any given task has the most context and information to determine where that line should be. They should be able to decide if they need 0 reviewers, 1 reviewer, 2 etc and who those reviewers should be. In general it's not been the most capable engineering teams i've worked on who've had a 1 review minimum. The most capable teams typically have the best test suites, style guides, feature flags, engineering practices and experience to trust developers to decide

I am not fully sure what part you don't follow. Are you saying that if I change the code, and someone reviews it, I have no reason to trust to that review either, based on my own logic? Is that what you're saying? If so, I don't buy that argument. Whatever you do, you always benefit from reviewing the results. Self-review is already beneficial, but if you can get someone else to do it, you increase the probability of discovering defects for two reasons (1) you become a bit blind to your own work and (2) other people may have knowledge you don't.

I tend to think these are largely orthogonal issues. People are imperfect (and so are agents), and they will make mistakes. You can have a high-trust team but still require human verification of code changes to build in quality and reduce defects.

@penberg Yeah I don’t really understand what you mean here because you could just apply the same argument to your review of someone else’s code or someone else’s judgement to review your own code and then you end up nowhere.

I absolutely don't trust people to predict ahead of time if their change has bad consequences or not. In fact, I don't trust myself either. That's pretty much just good engineering. That's what I meant: I believe (human) review as a method to improve quality is orthogonal to how much you trust or distrust the team.