Shreya

the last thing my bangers see before being posted

At #WTFbyMarmeto, the conversation got me thinking about ai tells that make me want to close my tab. It's sentences like this 👇 "It's not just evident - it's also highly obvious."

Since I posted this video, I've had three different people representing various groups of museums reach out to me asking to collaborate. The app took 2 hours to vibe-code on Sunday afternoon. We're so early / You can just do stuff / build in public

Me defending my O(n^3) solution to the coding interviewer.

Most valuable thing I learned from a senior engineer: How to read a codebase you've never seen. 1. Find where requests come in 2. Follow one path end to end 3. Map the data flow, ignore the logic 4. Only then zoom into the details Took them 10 minutes to teach. Saved me years of fumbling. Some skills are so fundamental we forget they need to be taught explicitly.

This is nothing. I LITERALLY created an agent to replace my entire family during the holidays: .claude/ - agents/ - daughters/ - snack-negotiator[.]md - why-asker[.]md - sibling-conflict-instigator[.]md - screen-time-lobbyist[.]md - spouse/ - calendar-synchronizer[.]md - dinner-decider[.]md - social-commitment-rememberer[.]md - gentle-reality-checker[.]md - parents/ - unsolicited-advice-generator[.]md - guilt-trip-scheduler[.]md - grandkid-spoiler[.]md

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

crazy how nobody is paying attention to how AI researchers made the most valuable self-development manual in history you can literally become more intelligent by copying AI research playbooks anywhere AI makes mistakes is also where humans often fail: - hallucinations: how do you know what you know is correct? have you tested it against the real world with real consequences if you're wrong? how are you evaluating? - context window: do you have a good understanding of the inputs, outputs, and constraints? how do you summarize a conversation to make sure you can transfer it to a different context? are you breaking things down step-by-step and making checklists to solve your problems better? - few-shot learning: how can you use examples to learn a pattern? what are the mental models you can extract, and how do you know they're right? - training data: you pay for every piece of content you consume. everything you put into your brain affects the quality of your mind. it also affects the quality of ALL outputs. you become what you consume: do more math and you'll start being more logical overall, not just when solving math problems - generalizability: can you take concepts from one domain and apply it to another? what are some of the most useful tricks you can take from an unrelated field and apply it to your own? can you still perform well when the context changes (time pressure, higher stakes, different culture, new feedback loops)? - prompt sensitivity: a small change in wording can change AI output. humans are just as volatile depending on framing, priming, and social pressure. think of "triggers." it's less about truth and more about cues that spark memories - overconfidence: if the AI sounds right, does it mean it's right? how many of us fall for charismatic "sounds right" if we're given a good enough story? - reinforcement learning from human feedback: AI models are pushed to be socially correct. how much of what you think has been affected by self-censor, virtue signaling, hedging, conformity, politeness, and reputational incentives? - multi-step reasoning failures: are you thinking too far in advance without verifying in between? humans lose correctness with multi-hop reasoning unless you get feedback and run checks - tools and functions: AI gets better with external support like search, calculator, and notes. you can also get better performance with systems, checklists, and environment design - prompt injecton: people can use social engineering, rhetoric, specific vocabulary to bypass guardrails and create errors. how are you making sure not to fall for the same manipulation, motivated reasoning, or authority? - benchmarks: when you optimize for a target, you'll get good at optimizing for its metrics. are you optimizing for grades, likes, promotions, status when your actual goal is something else? - system prompt: how are you looking at the world through the lens of identity, values, and defaults that you carry from culture, upbringing, and roles? and my favourite one: just because it can make sentences, doesn't mean it can reason (just like a lot of people you might know...) what else?

Career Alpha: Of all the most right tail outcome people I have met, a common thread tying them together is a confidence bordering on insane that they can achieve anything, learn anything, do anything. I believe it’s this specific trait that allows them to jump into the abyss for any task or project, no matter how gargantuan or seemingly impossible it is. Of course, most people won’t make it and achieve the right tail outcomes, so you can save your airplane memes. But without this confidence you never give yourself the lottery ticket of overachievement anyway.

Yup

A useful exercise in relating to emotions skilfully is treating them like animals in a documentary we are filming. We start by making ourselves very quiet and still so that they are undisturbed by our presence. We make ourselves alert so we can notice small details about how they act. Then, we investigate what they are and how they behave. Starting with a completely open mind and deep curiosity. If, for example, anger arises, we don’t just think “ah that’s anger” and assume it means we should do a bunch of angry actions. A good documentarian doesn’t participate in the dramas of the animals or become identified with their struggles. Likewise, with the emotions. Instead, we notice their shape and texture. We notice how they move. We notice how they interact with the other feelings and thoughts in our internal ecosystem. We let them express their nature without interfering or getting involved. Just as a documentarian may watch two lions fighting without going in to try and break the fight up. The more we get to know emotions in this way, the less they become our masters. Then our emotions can become like animals in a show we are directing, rather than making our lives a show that they direct.

I’ve spent the last year watching enterprises go from let’s automate tasks → let’s give agents business goals & let them figure it out. if you’re in #Hyderabad on Dec 6, come be a part of the #agenticworkflows conversation 👇👇👇Signup here community.automationanywhere.com/events/pathfin…

I genuinely do wonder if I get a different experience to most users, since coding agents know who I am

the Michelin man is kinda like a Greek god in the sense that he presides over tires and which restaurants are good

I have a pupil in my class (Year 6) who did this in about ten minutes. She's massively talented. Some ideas about how to help her move forward would be very much appreciated. Art isn't really my speciality.

Look, if we remove the slippery slope fallacy from the list of logical fallacies, then before you know it we’ll be removing all the other informal fallacies from the list and will soon forget the very concept of an informal fallacy altogether.

I consider myself a "just in time" developer. Like I have no idea how to work with sqlite databases in Python. At least not without an ORM. By the end of the day I will. Let me tell you the secret of how I do that... I read docs and play with code until it works. 🤫

Love it when art brings people together ✨ youtube.com/shorts/fC76H7G…

you really can decide to just stop believing in the concept of shame. that doesn't mean you'll stop feeling it, but you can just be like "yeah I'm over that"