perhaps my mornings alone or the way i prepare a snack of brie and crackers will have rhyme. isolated moments, sure. but i don’t search for consistency or a steady thrum in my life.

each day feels different. i want to live a life nobody can replicate, because i weave each day together solely by the string of my own curiosities and ambitions.

there’s a whole movement romanticizing routine: checkboxes, predictably slow evenings. nothing wrong with that, but i think it’s romantic to live a bit more freely. a day should be enriched with even the slightest bit of novelty - trying a yoga class, making tiramisu with a roommate for the first time, hopping onto a random podcast recording meeting … some of my recent shenanigans, and these don’t fit into the same 4-count.

some days have an 8-count. some don’t have a count at all. there’s no metronome to my weeks. certainly not my years. i want to work and sidequest without knowing my own trajectory a few years out.

despite our tendency to overarchetype everything, i don’t believe most of us are truly predictable. conforming is comfortable, and that’s what pulls us into the same few counts.

so i’ll play out of tune. out of rhythm. a misstep here. or there. a few moments dancing to someone else’s swing.

Subscribe now

today’s drops 🔎

• check out the paid mats research fellowship (housing & food provided)

• technical product manager internship @ amazon

• $150k for 30 teams building in hardware (apply by sept 30th)

my scope & goals

before browserbase, i had only worked at companies with <10 people. this was my first experience coordinating with a larger team while still taking ownership over projects of my own. my intentions this summer with this kind of dynamic were two-fold:

1. wedge myself into a team as a friend and a competent team member people can rely on.

2. become more technical and ship a lot.

at browserbase, i sought to immerse myself in the world of bringing more eyes to a technical business. this manifested through each of the projects i took care of.

largely, i had 3 main rocks at browserbase: writing technical content + blogs, developing a content engine for SEO/AEO, and improving the ceo’s (paul’s) linkedin presence through content + strategy.

my ships

BlogEO (my engineering project)

after a 1on1 over chai with jay, we both concluded that i should build something that continues to provide value even once the summer wrapped up. with some encouragement, a lot of scoping and debugging, and a final few autonomous tests, i built blogEO. it’s an engine that audits, rewrites, and generates SEO/AEO-tailored blogs on a weekly basis, and then measures if any of it worked.

the full write-up on the project and how it works can be found here.

the blogEO writeup

23 blogs + 5 CMS features

i wrote a lot of blogs. 29 to be exact. i won’t list them all out, but some favourites:

• build vs. buy: how to build agent infrastructure yourself → this was a sales enablement resource where i built a bunch of custom embeds. it ended up hitting front page on hacker news!

• the captcha arms race: from distorted text to browser identity → explaining the history of captchas and their progression through a mini-game i built.

• why we forked chromium → a technical explanation of why we built on top of playwright.

• how magicare turned a 60-minute decision into 60 seconds with browserbase → my first case study blog! i interviewed the ceo and wrote up a blog explaining their browserbase use case.

• how serverless browsers unlock scalable web automation → a deep dive explainer on serverless browsers.

you can check out the rest on the company blog site here.

i also shipped 5 features to our content management system (cms), like a stat card component, in-blog tables, embed iframes + auto-populating html, an unlisted toggle, a faq component, and cleaned-up code blocks.

my writing largely contributed to the following impact:

• 5.8x search impressions

• appearing for 9.8x page-one queries

• average position 10.9 → 6.6 (page two → page one)

• 6.8x pages earning impressions

250k impressions & 32 posts on paul’s linkedin

i also ghost-wrote for paul’s linkedin and developed a new lead-gen channel. i honestly had 0 experience in both of those things - i had no idea what lead-gen even meant.

basically, i took over paul’s linkedin, writing posts and tracking analytics for a specific ideal customer profile. i had a whole feedback loop set up with paul for idea generation and approval. i also messed around with sequences, signal-tracking and other business-y things.

i drove 250k impressions (+83% from prior) & 3.7k engagements through 32 new posts.

some miscellaneous ships:

• i helped launch browse.sh on product hunt

• shipped our internal mission statement with paul

• shipped ui for our internal content dashboard

main learnings

i first joined browserbase to write technical blogs on a contract basis. i was ecstatic to accept a summer internship offer to expand my scope beyond writing - to real engineering work that moves the needle of the business. i came to my manager, lindsay, with 3 big rocks in mind, and somehow my intentions converged into something greater.

i took ownership in a real engineering project for the first time of my life. i designed it, built it, tested it, shipped it, and iterated.

the world of devtools, agents, browsers, even writing code was pretty unfamiliar to me, and now i’ve learned so much in those areas + more.

i had the opportunity to fall in love with the “growth engineering” role. picking a problem and solving it across every function in a business, with the goal of driving eyes and revenue. even more, i absolutely loved the team. i made friends i plan to hang out with in the future, stayed up to push a launch with excitement, and drank a ton of brown sugar lattes and matcha during my one-on-one walks.

i learned, experienced, and grew a lot.

specifically, i learned…

how to build real processes

for something to have longevity, there needs to be a process for it, with other people involved and aware of what’s going on. i learned a lot about what these processes look like in other projects (even just reporting on ships & especially lead-gen/bdr stuff), and adopted these practices into my own. this looked like writing up one-pagers and scoping my blogEO project, reporting paul’s linkedin analytics on a weekly cadence, setting up a new slack channel for testing purposes, and more.

how to be a good team member

pretty cliché, but i learned what it meant to be a good team member. i learned from example and the amazing culture at browserbase and carved out time to have meaningful one-on-ones with people across the whole business, invested myself into the work, asked questions when i was confused, and did my best to help out others when they needed an extra hand. something that was slightly uncomfortable to me was also being loud about what i’m doing. tapping people in, writing slack messages, telling people and updating them on my progress - that’s normally where i’m pretty silent. i’ve spent most of my life working alone on things. communication is the most important step towards calibration.

all about devtools, agents, browsers… the list goes on

of course, i also obtained a lot of technical and subject-specific knowledge that i most definitely didn’t have before. my background is primarily in healthcare and education. i weirdly know a lot about medicaid billing codes, despite being canadian. stepping into browser agents, software infrastructure, growth as a role, and everything that comes with “ai b2b saas” (…) was completely new. i’ve enjoyed it a lot and have been able to draw a lot of parallels to other industries as well - many neurons firing at all times!

i wrote this blog on how browserbase felt like the goldilocks startup internship.

gratitude 🫶

this was an incredible summer, and it wouldn’t have been possible without lindsay and paul taking a bet and placing so much trust in me. culture is a hard one to crack and i’ve been able to end off the summer with so many genuine friendships - people i can trust and count on to both have a laugh together, or an idea to execute on.

what’s next

i’m back unc chapel hill now for my junior year of college. i’m excited to spend this semester continuring to write this publication (3+ years now!), exploring new problems and building side projects as solutions. i want to soak up my college years while still nurturing my curiosity. :)

Subscribe now

today’s drops 🔎

• apply to this full-ride scholarship to duke university or unc chapel hill (open to all high school seniors around the world)

• data engineering internship @ tesla

• 3 days left of this $35k webmcp challenge

sometimes the practice disappears altogether. sometimes it survives, while the reason doesn’t. maybe you still oil your hair, but you have no idea why.

knowledge, especially cultural knowledge, tends to be abstracted away with each generation. even broader, people know more now than 100 years ago for sure, but it’s new knowledge. the “old” stuff gets lost over time. this is a good thing when it comes to innovation and scientific or technological progress - a field moves forward partly by discarding what it doesn’t need anymore.

it’s a hard thing for culture, especially within the immigration context.

without a community actively upholding cultural traditions and language, it’s hard to continue funnelling that knowledge into future generations when it’s just you.

so if nobody tells you, then looking it up is the only path left. and looking it up has been getting narrower.

with books, hunting for an answer to one specific question would have given you a lot of other information and facts as well. that spread narrowed with search engines. and then further now with answer engines. you ask, it answers, and nothing else comes with it.

without the random exposure to new adjacent information, we lose breadth without noticing we’re losing it.

this is the problem, because you don’t know what you don’t know. you can’t search for a gap you can’t perceive, and nobody is left to volunteer it over. without intentionally seeking out complete gaps in our knowledge, we will lose much of the knowledge that unknowingly built our childhoods and culture.

and eventually someone will ask me why we do x or y. you cannot teach others what you do not understand yourself.

Subscribe now

today’s drops 🔎

• apply to the $55k telora fellowship by november 22nd

• project management internship @ american express

• sign up to be notified when true ventures fellowship applications open

his promise is already legal. the illegal part is a bit different, and that’s the main scoop.

how beef gets to a shelf

calf: 2 years on a ranch → feedlot, where it gets fattened up. safety rules don’t really apply at this point. that happens at the plant.

first, a government inspector checks every single animal while it’s still alive. the sick ones get pulled.

next, they’re killed and skinned, because the hide is covered in bacteria. most of the contamination that ends up in your beef is from this step. it’s the riskiest.

then, inspection #2, which was a rule put in place in 1906. for the next 90 years, it was a “poke and sniff” test.

next is the safety plan, with some history behind it.

in late 1992, e. coli in jack in the box burgers killed four kids and made 732 people sick, most of whom were less than 10 years old. how’d that slip through inspection #2? well, you can’t smell e. coli. poking and sniffing wasn’t gonna catch it.

so the government banned selling beef that contains it, even raw. and in 1996 it made every plant write out a plan naming every point where bacteria could get in and what they’d do about it.

then they chill the meat, cut it, label it, and ship it cold.

there is indeed a monopoloy

4 companies buy 85% of the cattle in america (36% in 1980). in 2022, 12 plants made almost half the beef in america.

this also isn’t a politically left or right thing. the bill that would actually fix this has both a republican and a democrat behind it.

but farmers can already do this lol

there’s been an exception since 1906. you can butcher your own animal for yourself, your family, and your guests. what you can’t do is sell it.

even if your state inspected it instead of the feds, you still can’t sell it across a state line, even though states have to be “at least equal to” federal standards.

so… the right to process your own food already exists. you just can’t sell your neighbor a steak. that nuance isn’t as easy to make a flashy tweet about i suppose.

the right wasn’t the hard part

one farmer built a fully approved butcher shop in 2019, and he still couldn’t open it. he couldn’t find anyone who knew how to butcher.

and that’s after the 2 year approval process, during an inspector shortage.

sure, delete a rule with a memo, but you still can’t abracadabra a butcher.

Subscribe now

today’s drops 🔎

• silicon valley fellowship applications are open

• product management internship @ datadog

• get paid to work on an open source project with sequoia

i would think, “you don’t know me … why are you being overly familiar?”

sure, sometimes they are putting up a front. but i’ve come around to thinking there’s genuinely no harm in greeting others warmly.

i realized that the suspicion was never really about the other person. it’s cheaper to assume someone’s performing than to match their energy. calling it fake let me off the hook.

greeting with warmth closes a gap of unfamiliarity more easily than small talk ever would.

there’s a way to do it with sincerity. warmth is a hug after a great conversation and asking questions with genuine enthusiasm. it’s not planning to hang out without any real intention to make it happen.

it’s more than just being friendly too. we’re probably the most socially awkward generation to become adults so far. everyone is waiting for someone else to go first, and warmth is just going first. so go first.

Subscribe now

today’s drops 🔎

• apply to this writing travel retreat by september 1st

• global operations internship @ wells fargo

• check out this job-fair styled event in nyc with reach capital

everyone is wearing carolina blue, people are walking around everywhere, and it’s beautiful, with trees so dense they pack the sky. it’s a college town in the way you intimiately get to know every single street by their memories and the last person you walked into a coffee shop or 2am burrito spot with.

there’s belonging in the feeling that this town is meant for you and your group alone.

it’s unlike the experience you’ll get in a big city. you can also get a big city experience any time in your life. and you probably will, if that’s where your career aspirations take you.

the college town experience is uniquely offered only within the college timeline. every year, i’m more grateful for the choice i made and how much i’ve grown to adore chapel hill.

it’s a college town in every way possible.

Subscribe now

today’s drops 🔎

• apply to the $55k telora fellowship by november 22nd

• project management internship @ american express

• sign up to be notified when true ventures fellowship applications open

once i get to the restaurant, perhaps my plan will change based on an employee’s recommendation or the look of a different dish on another table. but i can only clock that the recommendation is better because i already read the menu. otherwise i’d just nod and order what i came for.

regardless, i still know the menu. i’m okay with changing my mind. i probably have a question or two in mind as well.

i prepare for my coffee chats in the same way.

i get to know enough about the person from their twitter or linkedin. maybe they have a few blogs. i’m interested in chatting with this person for a reason, so i’ll just learn more about their experiences or thoughts within that realm.

i will also make sure that i have an intention in mind. this is the most important part. i make those intentions obvious through my chats. if i’m chatting with a founder and i want to work at their company, i’ll tell them straight with a quick pitch on why i make sense to bring on board.

i convey my intentions, but i also have my own curiosities. i come with some questions in mind, but they aren’t a checklist. i ask as the conversation flows and my curiosity remains. the initial research is what lets me hear when something unexpected gets shared - i know it’s unexpected, so i can chase it.

i can’t blindly order something on the menu without a photo, description, a review, or perhaps someone’s opinion.

you can never overengineer a chat. but you can prepare around your intentions for it.

Subscribe now

today’s drops 🔎

• hacknc registration is open, unc’s hackathon from october 9-11

• finance internship @ etched

• apply to dorm room fund as a student venture partner

i’ve realized that i do enjoy cooking. it’s quite a leisurely hobby, though. cooking something elaborate realistically takes a lot longer than the sole act of cooking.

shoutout to my mom because i’ve only now realized as an adult how time-consuming it is.

cooking means planning for groceries accordingly while keeping existing produce and dairy that’s closer to going bad in the fridge in mind. getting the groceries. actually cooking. minimum half a day spent on a new recipe? idk about that one.

even if this is something i enjoy doing, i doubt i’d ever choose it over the millions of other things i need to do. if i had nothing to do and a stocked kitchen already ready for me, sign me up.

so i meal prep. i like to eat more than i like to cook.

the cooking itself isn’t that time-expensive. the planning and the dishes are, and they don’t scale with servings. so one grocery trip and one cleanup can cover five meals just as easily as one.

with meal prepping, i can at least drop the quota from 1 cool new meal a day to 1 a week.

i still make the same rice bowl every dinner … but i figured i should vary it a little.

some meals i want to try prepping in september (yes all i do is save food on tiktok):

1. cottage cheese bhurji with toast / roti → a spin on the classic indian paneer bhurji

   1. also chili cottage cheese toast (i just want paneer but make it protein)

2. corn & feta scoopable salad with chips → to snack on….

3. mayak eggs → they’re super versatile to just eat with rice, ramen etc.

4. vodka pasta with tofu blended into the sauce → classic tiktok meal.. don’t judge me

   1. i also want to mix in a bunch of indian flavours, because food is too plain without it i fear

5. paneer tikka tacos → just prepping the paneer + chutney with it

so … i obviously like indian food. shoot me. i’m going to try prepping 1 new meal every week to keep my life more interesting. i’m slowly getting tired of my beloved tofu rice bowl and have been itching to cook more creatively in my chapel hill house.

prep endeavours coming soon.

Subscribe now

today’s drops 🔎

• contrary student venture partner applications are open

• business transformations consultant internship @ ibm

• check out this job-fair styled event in nyc with reach capital

i like reading about historical innovation and joking about how i could’ve discovered x thing as well. it’s an easy cop out to say that there are fewer ideas left to research and invent now.

there are still a plethora of ideas. though the barrier to entry is probably higher now.

it’s higher because the frontier has moved. newton needed calculus and an apple; meanwhile today you’ll need 10 years of prerequisites before you can conclude anything new in the same field. you can’t count on an apple falling on your head and making a groundbreaking discovery anymore.

ideas are born from exposure and reflection.

if you don’t have enough ideas, you need to read. if you don’t know what to make of your reading, you need to talk to more new people.

an idea demands novelty, and that means you must expose yourself to a fresh topic, conversation, or new author. if the barrier to entry has risen on new ideas, then raise your knowledge alongside it.

ai has made that part table stakes. you can get up to speed on almost anything now. so there’s no real excuse to flip over a table and conclude that no ideas remain in the world.

ideas come from knowing or observing something that other people don’t. if you know the same things that everyone else knows, what could you possibly think up that’s new?

read more. have more conversations. ask a thousand questions. and then write down what you make of it.

Subscribe now

today’s drops 🔎

• silicon valley fellowship applications are open

• risk & security internship @ vanguard

• get paid to work on an open source project with sequoia

i like to be reminded of my life. it grounds me.

i take a lot of photos. many things are beautiful to me, and even more things are worth a story. to recognize this, one must also practice gratitude.

you spent $200 to see bruno mars, and every single person has their phone out. it’s a classic modern scene. it’s funny to laugh at, but i don’t have a personal problem with others having their phones out to record a memory. it’s not my life.

criticizing others for not living in the moment is an example of people feeling weirdly philosophically superior. sure, most people are likely doing it to post on social media, but we all have our own attachment to individual memories and experiences.

if it makes someone happy, why not?

my personal stance actually leans more pro-photos than neutral.

the memory space for stories at the top of your mind is honestly very small. you can only recall a few stories right away without any sort of prompting (a conversation, a photo, someone’s earrings) - photos are both kindling for stories and relics of your memories to look back on later in life.

perhaps it’s why people (me) love to go through photos and tell stories as they scroll. it’s fodder for your narrative.

i wonder what a digital archaeologist will make of my life from my camera roll?

i think if you can be honest with yourself when asking, “am i taking these photos to capture a memory, or to get a perfect social post for the experience?” then you’ll know whether taking photos is taking away from the present.

don’t get me wrong, i love getting the right photo to curate onto an instagram feed, but if it’s taking away from an actual meaningful experience that i’m supposed to be having, then i’ll take a quick photo and keep it pushing.

it’s different if i specifically invite my friend to wear a cute outfit and go to a café to take pictures. i’m not going to spend an hour taking photos if i’m hiking in yosemite. those will be captured candidly or through a couple of poses with friends at the summit, and the experience will continue.

i take photos of everything. i have a pretty thorough documentation of my life from maybe 11th grade onwards. photos and these blog posts remind me of the stories that make up the moments that i did indeed live while being present.

we exaggerate how much a photo takes away from the moment. it boils down to your intentions.

Subscribe now

today’s drops 🔎

• check out this $1000 game theory event sponsored by jane street (apply by sept 1st)

• software engineering internship @ figma

• 3 weeks left of this $40k agents hackathon

halvdan.

the runes are worn down now, but enough survived the test of time to read this name. halvdan was scandinavian - probably one of the northern men who travelled east and ended up in a byzantine emperor’s military service. further along, there’s a second inscription that reads something like “arinbarðr carved these runes.”

he was definitely bored, because you’d end up in the gallery during a very long service. the man has marble and a knife right in front of him, why not? i’d like to imagine it was his version of us doodling in our notebooks in school while our teachers would drone on and on.

this inscription doesn’t tell us anything about his job. we don’t know why he came, or whether he went home. it’s simply proof of his existence and that he was in that room.

a thousand years later, it’s still legible, and we’re still talking about him.

we uncover ordinary people all the time. they’re the fabric of a society, not the kings or generals. around 1750 bc, a man named nanni was sold some bad copper and wrote an angry letter about it on a small clay tablet. that tablet lives in the british museum now and has outlived most of the literature of the city it was written in. a customer complaint.

around ad 100, a woman named claudia severa sent her friend a birthday party invitation written on a thin sheet of wood. a scribe did most of the writing, but claudia signed it off with my dearest soul, which is likely the earliest surviving handwriting by a woman anywhere in the roman world. a party invite.

this isn’t the legacy we think of and aspire to leave. it’s a trivial week, yet still recorded and resilient through time. we have it because the world used to be physical, and creation took up real space.

physicality is the closest thing we have to permanence, only really vulnerable to natural disaster and time.

our world now, the digital one, is vulnerable to much more.

now you can subscribe to permanence monthly!

physical things survive by being abandoned. archaeology is pretty much the study of what neglect managed to preserve. halvdan’s runes lasted a thousand years because nobody sanded off the marble.

digital things, on the other hand, survive by being maintained. someone has to care for it by renewing the domain, or migrating the database when the cms deprecates.

digital permanence requires a living person to keep caring for it. but people stop caring. almost always. unless you want to automate anything and everything in the world and we’re just a world of bots.

pew did a study in 2024 and found that 25% of all webpages that existed from 2013-2023 were already gone. from the 2013 pages, 38% vanished completely, and 8% of pages from 2023 (less than a year old then) failed. more than 50% of wikipedia’s reference sections have at least one dead link. brewster kahle, founder of the internet archive, even estimates that the average lifespan of a webpage is 40-100 days.

i literally dealt with this at a smaller scale this summer during my internship. one of the functions of blogEO’s (the tool i built) is to audit our blog for broken links and fix them. a new broken link pops up and gets flagged every other week.

digital rot doesn’t leave rubble behind.

when a temple collapses, the pile of stones left over is the record. but when myspace lost 12 years of uploaded music in a server migration (~50 million songs from 14 million artists), nothing was left behind. all gone, in one fell swoop. this user wrote that they lost 7 years of diary entries, travel anecdotes, and memories of their father, who had since passed away. they lost 7 years from their 20s.

We thought it would be there, or that they would give some warning and a way to retrieve data if they suddenly decided to gut everything. I just lost what amounts to 7 years of a diary, travel stories, reflections, memories of my (now dead) father, 7 years of my 20s just vanished, gone.. It’s not upsetting, it’s devastating. I’m grief stricken... silly me, thinking my content was mine.It’s not even like a hard drive crash, a hard drive can be easily backed up but blogs were notoriously hard to, maybe purposely hard to. And now..all gone.

digital gone is much more complete than the physical gone.

the forgotten century

vint cerf, one of the actual architects of the internet, has been warning about bit rot since 2015. he has said that we’re at risk of a “forgotten generation, or even a forgotten century.”

even if a file survives, his point is that the thing that can read the file might not. maybe in 100 years, a perfectly intact document will be retrieved, but it’s a useless brick if it’s in a format that nothing opens anymore. each software update migrates entirely towards the next layer of efficiency. cerf proposes digital vellum, which is preserving the data, yes, but also the software, hardware, and operating system.

we’re the first generation with this problem, inventing a plan for how our writing will be read. we didn’t have to build a reader for marble.

we’ve started building mountains

the digital world has become so important that we’ve begun to invest in its permanence.

we have backup generators and redundant power, failover regions and offsite replicas, and we put cellular data and starlink on planes because 8 hours of disconnection is utterly unacceptable…

we’re engineering around the digital world’s fragility. we do this for the physical world too when we engineer around famine and flood.

on february 2, 2020, github took a snapshot of every active public repository and put it on film. it feels like something out of a sci-fi movie, but 186 reels of film were encoded by a norwegian company, rated for at least 500 years (tested to suggest 1000 years), and sent to svalbard, which is closer to the north pole than the arctic circle.

it got buried in a coal mine under hundreds of meters of permafrost.

the data is qr-encoded with a human-readable index explaining how to recover it.

wtf. the irony is incredible. to make our digital artifact permanent, we converted it back to physical and literally buried it in a mountain. we haven’t found a digital solution to digital impermanence, so we went and found a rock.

your 2019 “fix typo” commit is sitting in the arctic right now. it’ll live longer than both of us.

so what will be found of me?

i’ve been publishing here every day for a little over 3 years now. that’s ~1090 posts. the oldest thing i have online is either my first beehiiv post from august 1, 2023, or my instagram account that i’ve had since 2017. perhaps there’s also something older that isn’t yet coming to mind.

this could all just vanish. i was already nervous when i migrated this publication from beehiiv → substack, because neither of these companies owes me anything. if substack goes, then i’ll export everything and put it on my own hard drive. it can still all disappear, but some plan feels better than no plan. hard drives still fail. and i will also die.

i’ve already experienced this sort of loss. i used to run an organization in high school that taught programming to kids in underserved neighbourhoods. when it wound down, i deactivated the domain, and all the curriculum, lesson materials, slides, worksheets, and everything we built in the google drive disappeared with it.

i can describe them to you. but i can’t show them to anyone.

there’s no rubble.

on the other end of the spectrum, i went on a 3-week sailing trip in maine 2 years ago with no technology. my mom sent me off with an envelope of 21 index cards, with one note for each day. we sailed through some treacherous conditions (hurricane included), and i still have every single card to this day. in the envelope. completely fine.

we now consume more, but somehow it will leave less

i was very quick to think that consumption leaves nothing behind, and creation leaves behind everything, but it’s actually worse than that.

two researchers, sarah gabbott and jan zalasiewicz, wrote a book called discarded on “technofossils.” the main argument is that a flurry of new objects have appeared on earth in a geological instant. within a few decades, we’ve produced tonnes of plastic bottles, pens, concrete, teabags, phones, t-shirts .. and the list goes on. most of these are also specifically engineered to resist the natural elements, piling up in landfills in quantities that outweigh our whole living world.

much of this will petrify and become a legible layer in the rock.

so consumption in fact leaves an enormous amount behind. this laptop i’m typing on will plausibly outlive every word i’ve ever published. my plastic target bag may very well be in a sediment layer long after the servers holding my 1090 posts have been decommissioned, resold, and melted down!

we’ll leave behind a lot, but it’s all meaningless. the difference is an attributable versus anonymous paper trail. my landfill will tell a future geologist about the type of person i was. my category or demographic within my decade. some broad labels.

none of my stories. halvdan left behind a name, a scratch in a railing.

he certainly had more durable physical records like coins and bones. but that record is anonymous. the scratch was fragile. the fragility is specific.

waiting to be found

i don’t care much for legacy.

i care about how deeply i’ve wedged myself into the world i was living in. i want there to be evidence that i was in the room. the withered stories on papyrus and clay have become scattered blog posts and emails in random corners of the internet on infrastructure we don’t own.

they do exist. i want those individual stories to continue existing.

and in 100 years, by some slim chance, maybe some of these pages will still resolve. maybe someone will find a cached copy of yesterday’s post and accidentally get a snapshot of a specific college student in a specific year. they’ll learn the fashion, the music, or how a regular week panned out.

perhaps a future digital archaeologist will invent a ritual purpose in my camera roll, the same way we do for figurines we don’t totally understand.

in fact, yale has run a digital archaeology and preservation lab since 2015. archives new zealand even bought special hardware to pull raw bitstreams from 5.25-inch floppy disks that no ordinary machines can read anymore. it’s the same job, split into digital archaeology as well now.

there are people excavating soil and people excavating disks. eventually, there will be more of the latter.

halvdan didn’t archive anything. a knife, a wall, and some spare time ended up being a more resilient publishing strategy than mine.

i write every day, partly because i think you forget your own life if you don’t, and partly because i don’t think you can think without writing. i’ve never really written for the distant future, but i truly marvel at the fact that a bored man’s name outlasted an entire empire.

so what’s the paper trail? what did any of us actually carve?

and in a hundred years, will an archaeologist find any of it?

Subscribe now

today’s drops 🔎

• check out these archaeology research grants & awards

• merchandising internship @ louis vuitton

• apply to this $10,000 fellowship to solve an ambitious problem

tl;dr:

• the browserbase blog had no SEO/AEO strategy or tooling → 53% of our search traffic came from 5 posts, ~74 posts had empty SEO fields (title + description), and only 17/77 posts had ever been cited by an AI engine.

• i built blogEO, a weekly engine that 1) audited existing blog posts and suggested fixes, 2) generated new blogs to fill real search demand, and 3) tracked AEO metrics.

• importantly, the agent (bb) has no write tool. it only scores, drafts, and posts buttons into Slack. a server-side handler does the writing after a human clicks approve.

• each approved audit edit takes a before-snapshot along with the blog-wide numbers for the same dates. the snapshot gets re-read at 28 and 56 days.

• as of august 21, 2026, blogEO has generated 18 blog posts and 50+ edits.

a blog without measurement

developers read blogs. they’re a source of truth, and llms pull from them to build confidence about how to write code. if we’re not showing up, then we aren’t capturing the kind of traffic that compounds passively while asleep.

our blog was manual, unmaintained, and unmeasured. i knew it well - i had written dozens of browserbase blogs myself since october 2025 after all. the older posts referenced outdated product details, underperforming posts were untouched because nothing was monitoring rankings, and all of these analytics were spread across 5 places that nobody signed into.

when i pulled this data at the start of the summer, i realized this was an issue that needed a lot more than some scrubbing. a handful of posts on our blog carried the bulk of our numbers, while the rest were inert.

ranking but no clicking

the more frustrating part was the fact that we were indeed ranking. our informational blogs were ranking at positions 6-13, but weren’t getting any clicks. one post had 329,454 impressions at an average position of 9.4, but produced 857 clicks. that’s a 0.26% click rate…

google was showing us to 300k+ people, and basically none of them came through.

btw, position refers to where a link sits on a google search result page. position 1 means it’s the first link, on the top.

invisible to LLMs

i also wanted to know whether or not answer engines were quoting us at all. the AEO strategy would follow (i explained AEO here). out of all of our 77 blog posts published before june, only 17 had ever been cited. even that was skewed, with 3 posts doing most of the heavy-lifting.

why no solution?

each of these issues was manually discoverable. it was a pretty rough discovery process though - it meant someone had to bridge together 5 different systems (our CMS Sanity, Google Search Console, SEMrush, PostHog, and last week’s numbers) that didn’t integrate, and then redo that connection every week so the data wasn’t stale.

what i built: blogEO

i connected all of these pieces + some more. blogEO is an audit, generator, and AEO tracker. all three run automatically on a weekly cadence through our internal agent, bb, into one Slack channel. all three functionalities can also be called on demand through Slack.

1. the blog audit (fixes what exists) → it scores every published post based on opportunity and decay, drafts surgical fixes for the posts with highest opportunity, and posts each edit as Approve/Edit/Skip cards in a Slack thread.

2. the blog generator (creates what’s missing) → it goes through live search data for demand that we don’t match with blog posts, writes a draft with code snippets traced to our docs, and gates it on 4 automated checks.

3. AEO tracking → sees whether ChatGPT, Claude and Google’s AI Overviews quote us and determines why a post isn’t cited.

one core design decision

bb can look, think, and suggest, but i didn’t give it any write tool. it can’t change the website. the code that can runs on the server when a human clicks a button (approve).

this way, the worst a bad run could do was make a bad suggestion in Slack.

this design also made the code correctly malleable. the agent’s judgement lives in a skill file (cheap to change), while the write path is code with tests, and it revalidates everything claimed by the agent.

picking what to fix (audit)

the audit is one run: it scores every published post, drafts fixes for the highest-opportunity ones, and then posts the edits to Slack.

then the run ends.

the button clicks are handled later by a separate handler, and independently from the agent’s session.

the audit steps:

1. 5 batched data pulls from Sanity, GSC, SEMrush, PostHog & last week’s run (done through bb via tools).

2. hygiene scan across every post for broken links, dead images/embeds, positioning drift, missing SEO fields, and typos (done through bb & in memory).

3. score and rank the opportunity for each post, then verify factual claims in the top ~15 flagged posts only (done through bb).

4. draft surgical edits and mark two safe edit types for auto-publish (bb proposes and server validates)

5. persist each suggestion to KV and then a run snapshot to Postgres (done through server).

6. post the audit summary + threaded card per flagged post/edit (done through server).

scoring by opportunity

the audit ranks by headroom, not age. i did this because i wanted to act based on the answer for: “if i improved this blog post, how many more visitors would we actually get?”

to do this, each post is scored 3 ways against the previous 28 days. the biggest of these 3 estimates “wins” and thus decides what kind of fix to write.

1. recover → these are clicks lost vs. the previous window. therefore, a regression to fix.

2. ctr → the page-one clicks that our blog post didn’t get, compared to the expected CTR its position should’ve earned. therefore, a title/description fix.

3. rank → the # of clicks a page-two post would gain by landing on page-one instead. therefore, a content push.

the CTR lever needs a model of how many clicks x position should earn. that’s an organic click-through curve.

export function expectedCtr(position: number): number {
  if (position <= 0)  return 0;
  if (position <= 1)  return 0.28;
  if (position <= 2)  return 0.15;
  if (position <= 3)  return 0.10;
  if (position <= 5)  return 0.06;
  if (position <= 7)  return 0.04;
  if (position <= 10) return 0.025;
  return 0.01;
}

if i put this post through the CTR lever, i get a number for "opportunity” that puts it at the top of the audit queue.

~7k clicks of headroom. the value of “opportunity” > refreshing because “this post is old.”

the limitation here is that the 2.5% benchmark uses average position, which GSC blends across every query that a page appears for. it over-counts the real headroom - so it’s a soft absolute signal, but valuable for ordering in this audit queue.

two guardrails for ranking on the edit queue:

1. if impressions are low to begin with, a low ctr won’t score as high opportunity. no edit can fix the fact that “nobody is searching for this.” these kinds of posts are just marked as low-visibility and dropped from the ranking.

2. a real click loss > any estimates. a post that genuinely lost traffic will surface regardless of where it is in the opportunity ranking. this drop is determined if there’s both a material absolute drop in clicks + a real proportional one.

rationing checks over tokenmaxxing

the cheap checks run on all published posts (age, empty SEO fields, dead links, traffic drops, and a word search for old positioning).

the more expensive check (comparing posts against our docs) costs a lot of tokens, so it’s only run on the top 15 flagged posts + 3 more posts as a buffer, so the posts with low impressions aren’t left unverified forever.

these flags are also not from memory. the real page must be loaded, only direct contradictions are flagged, and the source is quoted

the edits are intentionally small

a suggestion can only change the SEO title/description, or swap one exact phrase/link. if a fix can’t be one clean swap, then the card shows up only with Edit and Skip. a human has to write in the edit in this case.

for public-facing prose, i believe this to be the right place to draw the line.

also, dead links with a known replacement + filling an empty SEO field are two classes of edits that are auto-applied, so they bypass human approval completely.

measuring whether it worked

a GSC snapshot from 28 days prior is taken once a real edit goes live. data on how the whole blog was doing over those same dates is also captured, because without this piece, a 20% improvement would tell me nothing since Google may have lifted everything.

writing new posts (generator)

the audit was actually v0 of my project. it set up the groundwork for my real goal, which was to increase SEO/AEO traffic. an audit alone can’t fix a blog where 72/77 posts get no traffic - the traffic isn’t even there.

the richest source of traffic is near-misses, where Google already shows us 150+ times, but we’re at positions 5-20 without a dedicated post. one of my runs on GSC found 120 of these near-misses, including “what is a captcha solver?” at 19.7k impressions and position 10.6 without us even having a specific blog for this query.

one generator request produces a publish-ready draft, unlisted, and waiting for human approval through Slack. same as the audit, bb will never publish the blog.

the generator steps:

1. pick a topic from GSC data or take a provided one (via seo-strategy skill).

2. write the draft with all code snippets pulled exactly from our docs (via browserbase-writing skill).

3. check draft against the writing checklist (via bb)

4. gate draft on 4 automated checks before saving anything (via code)

5. create the unlisted draft (via code)

6. post the Slack card with Approve and Discard buttons (via code)

7. clicking Approve publishes it unlisted, and starts measuring (via server)

3 skills

i separated out 3 skills for this flow in order for it to be maintainable, since 2 copies of a rule become 2 different rules and things get messy. if the pipeline file ever starts shaping what good writing is, that means it’s in the wrong file.

the automated gate

before reaching Sanity, a draft has to pass 4 automated checks. this way a failing draft doesn’t end up as junk to clean up.

1. strategy → ensures no banned terms and the post maps to a real content cluster + persona

2. structure → ensures a post opens with a labelled TL;DR section + a few other nits.

3. code provenance → ensures every snippet has the docs URL it came from. no invented code.

4. cannibalization → ensures that none of the target queries are already covered by one of our own posts.

   • writing a 2nd post splits our own ranking, so bb is forbidden from retrying/rephrasing. instead, the post that owns this query is identified and handed to the audit as an edit.

input.body.forEach((block, i) => {
  if (block.type !== 'code') return;

  if (!block.docsUrl) {
    failures.push(`code block ${i} (${block.language}) has no docsUrl (source of truth)`);
  } else if (!/^https?:\/\//i.test(block.docsUrl)) {
    failures.push(`code block ${i} docsUrl is not an http(s) URL: "${block.docsUrl}"`);
  }

  // advisory only, won't block the draft
  if (!block.version) advisories.push(`code block ${i} has no version pin`);
});

heavily gating prose and structure means hardcoding taste and that’s incredibly unmaintainable code. whether that writing is any good is the checklist’s job, which is done by bb before gating.

bb has 2 attempts to redraft on failure and then stops with no draft produced.

measuring a post that starts at zero

the edits had a “before.” a new post doesn’t, so it’s measured at a growth curve instead, with the same structure of 28 & 56-day readings which include the blog-wide numbers for the same dates. on approve, the post is published, enrolled, then recorded. this way, a recording failure won’t block enrollment → enrollment failure won’t block publishing.

whether LLMs quote us

GSC tells me if Google sends people, but i still had 0 data on referrals/quotes from ChatGPT, Claude or AI Overviews. the first thing i did was split “AI referral” into 3 segments:

• does an engine crawl the page?

• does it cite us in an answer?

• does anyone click through?

i shipped cite through a parser i wrote for the SEMrush AI visibility export, which is unfortunately export-only with no API. this maps each row back to a post slug.

click already worked, because Google folds AI Overviews into normal search performance, and the other engines classify by referrer domain.

interestingly, the posts that win in Google and the ones that get cited by AI aren’t the same posts. our strongest search performers were only cited twice by AI.

i ended up deferring the crawl layer because it needed a rescoping to access Vercel log drains behind SSO. i wrote up next steps in a hand-off document for my team.

how it all flows

put together, bb reads and reasons, one server writes, and they both meet a person in Slack.

the measurement lives in a schema across 5 tables:

• blog_audit_runs → scored snapshot of each post

• blog_audit_edits → what changed, before snapshot, +28d/+56d readings & cooldown

• blog_generated_posts → quality-gate report & checkpoints

• blog_aeo_signals → AI citation counts

• blog_aeo_domain → domain-wide AI visibility

three messages per week

i firmly believe new tools should integrate with existing workflows, not give people another website or login. fragmentation here is exactly what that looks like when it goes wrong. BlogEO’s whole interface is a Slack channel with scheduled runs:

• new blog draft → 9am PT Monday → the week’s generated post with Approve card + a note on what was written and why.

• blog audit → 9am PT Tuesday → a ranked thread of existing posts, hygiene findings, and Approve/Edit/Skip cards for each suggested edit.

• performance & tuning → 9am PT Friday → a report including growth curves, edit-measurement state, AEO diagnosis per post, and suggested measurement edits.

a card begins as a suggestion and ends as a link to a live post. the in-between states took the most refining to get right. the suggestion locks as soon as someone clicks, so two people can’t both publish. the publish will also refuse if someone has an unsaved draft of that post on Sanity, because that draft would be from an older version and would undo the edit when someone saves it.

outside of the weekly cadence, the same pieces are still callable:

• “@bb what’s the SEO performance of this blog: ___”

• “@bb draft a blog explaining how to browse the web securely.”

• “@bb draft a blog”

• “@bb give me a title for this blog: ____”

• “@bb run the blog audit”

our internal agent now also has a completely new toolkit.

btw, a requested topic isn’t a pre-approved topic - it still gets classified, checked against guardrails, and run through the cannibalization check. if someone’s idea fails, bb will stop and ask follow-up questions.

the results so far

18 posts have been generated and published, ~74 posts backfilled with SEO titles/descriptions, 53+ older posts edited since July, and 6 weeks of BlogEO running on its own cadence.

the most rewarding part has been seeing the team continue to use the tool and generate/approve more posts after i handed it off and completed my internship.

my blog initiatives at browserbase (both writing and this project) led to search impressions growing 5.8x, page-one queries growing 9.8x, and our average blog position going from 10.9 → 6.6.

a few limitations include the opportunity estimate being soft, crawl being deferred, rationing the fact-checking, and AI citation data being a manual CSV export for now.

what i learned

overall, i learned a ton from executing my first engineering project end to end. this was just one of the many things i worked on over the summer - i’ll be reflecting and recapping the rest soon.

some quick lessons i’ll expand on in a later post:

• i learned how to scope and rescope an engineering project with extensive planning.

• it’s important to be meticulous about measurement. controls are important!

• the ux is the product, so it has to exist where work is already happening. nobody wants another login.

special shoutout also to jay for being a great mentor, poking holes at my thinking, putting me onto the best chai in san francisco, and reviewing all of my prs …

onto the next! 🔨

Subscribe now

today’s drops 🔎

• apply to the contrary research fellowship

• software engineering internship @ nvidia

• applications are open for this full-ride scholarship to duke or unc (high school seniors)

if you made it all the way down here, you’re the goat.

a discussion question was raised: how are race, gender, age, sexuality, class, and ability similar and different? and then further, when thinking about sexism, homophobia, classism, and ableism, which is most similar to ageism?

there is a lot of overlap between ableism and ageism of old age specifically. uniquely, ageism is the one “ism” that has traffic coming in from all sorts of directions - as is expected when it’s all so relative. older folks will discriminate against youth for whatever reason and vice versa. there’s no fixed marginalized group.

the interesting overlap between the two is that i would argue that a subset of older ageism is simply ableism. capacities do change with age - but how much that costs you depends on what the world is built to accommodate. this tends to be the case for a lot of older discrimination, when seniors are deemed “useless” or vulnerable. as such, the world also becomes significantly less accessible, the same way it is for people with disabilities.

it’s harder to travel from one side of town to the other, there’s fewer people around who know how to accommodate, and the wider emotion inflicted upon people is pity rather than disdain. pity dominates these two forms of discrimination, while it only occasionally shows up across other identity labels.

ableism and ageism are also both types of discrimination that just happen to people. excluding disabilities from birth, someone could have spent their whole life free of any discrimination or inaccessibility, but an accident could change that. similarly, people grow into ageism.

that part has one distinction that sets ageism apart from everything else. it’s inevitable. it’s an inevitable form of discrimination that nobody is spared from.

we can use as many labels for identity as we want to segment a population, but we all converge at old age.

Subscribe now

today’s drops 🔎

• apply early to the a16z speedrun alpha fellowship

• technology internship @ humana

• check out the ea global conference in nyc from oct 16-18

if you actually use the product you’re trying to sell to other people, that’s the best initial signal you can accrue.

when i was connecting all of our seo tools and developing a methodology to track aeo, i was getting frustrated by the lack of apis provided by multiple tools. i couldn’t pull some vital data without manually exporting CSVs and importing them into a database. chopped. even for tracking aeo citations, it would’ve been easier for me to run an automation suite of queries across various llms to see how we are getting cited, instead of piecing together utms and vercel logs. and that suite would need browser automation too, since answer engines don’t hand me an api for what consumers actually see.

i needed apis for websites that don’t have apis.

that’s the exact problem statement browserbase solves. i thought it was pretty funny and satisfying to think, “i wish everything just had an api bruh.” it solidified my own conviction that the product-market fit is real.

even me, someone building a project with very little previous engineering background, ran into this issue from the get-go. so of course it’s a major use case at larger scales as well.

it’s not enough to say you would use it in theory. if you are part of your own target demographic, whether or not you actually use your tool is a good indicator of something worth building.

Subscribe now

today’s drops 🔎

• apply to the contrary research fellowship

• software engineering internship @ nvidia

• applications are open for this full-ride scholarship to duke or unc (high school seniors)

i normally switch one or two things around, but i’m really excited this time around. all of my professors have good reviews and seem friendly enough. the syllabi are also not particularly daunting. the readings seem really interesting.

i’m taking four classes this semester. i’m not overloading my schedule at all - i’d like to spend my extra time thinking through and building projects on the side.

two of my classes work toward completing my cs major requirements: a class on computer organization (microprocessors and assembly programming) and an introduction to the theory of computation. they seem interesting enough - i’ve been recently intrigued by hardware, so the conceptual connections through computer org could serve me well.

the next two classes i took out of some interest, but mostly to fulfill my gen-ed requirements that unc calls focus capacities. one is an ancient greek philosophy class and the other is titled “bad chinese aging.”

hear me out.

when we went through the syllabi today, i realized that the philosophy class is actually more centered around the philosophy of love, family, and desire. completely out of my “career” depth, and i love it - it’s very much aligned with my goal of becoming a worldly person who knows at least a little about a lot of different things. we had some interesting discussions around happiness and emotional relativity right from the get-go. i was also pretty engaged with my professor’s style of talking, which is an incredibly good and rare sign for me. i tend to tune out when people talk for a long time. oops.

same deal with the chinese aging class. my professor is super funny and we already had a lot of great discussions on how our society isn’t built to accommodate aging, the specific connotations around family, aging gracefully, and destigmatizing solitude specifically in asian and chinese-speaking societies. there are a lot of parallels to indian culture, and i spent last semester literally visiting the countries that are the subject of this class, so i feel particularly connected.

i only very briefly skimmed the titles and summaries of all of the readings, but it’s all insightful. best of all, it’s a fantastic forcing function for me to think about different topics, consume literature, and connect new ideas and insights to write about here.

i feel like i’m checking off so many boxes beyond just credit fulfillment. this is my first semester at unc where i actually feel excited to take some of my classes. good riddance to the first two years of smashing out mandatory prerequisites.

Subscribe now

today’s drops 🔎

• check out the paragon policy fellowship for college students

• software engineering internship @ american express

• check out this investment readiness bootcamp for high impact startups

with this dynamic, developers end up being the least loyal customers.

there are thousands of saas companies, each with self-serve options, but the more successful plays end up being b2b (the ai b2b saas jokes have some truth to them). developers also have the ability and ego to consider building a tool themselves. as soon as there’s some friction with the product, or a better option presents itself, it’s instant churn.

to succeed as a software company, it’s important to secure enterprise over individual customers if you want to see profit, and profit that grows. individual developers aren’t the end destination, but can often be the door to expanding into their employer’s contract.

of course there are outliers, where a particularly new idea dramatically shifts individual workflows (ex. chatgpt). even those outliers had their winning streaks, until competitors came up and once again fragmented attention and dollars (claude, gemini, etc.).

developers aren’t fickle though. this “developer disposition” is the result of being informed. they have the background to understand how these tools are built, and they’re in the know about what’s being launched every single day. other customers are “sticky” because they can’t rebuild the tool, don’t see the alternatives daily, and perhaps they did sign a bigger contract that’s hard to offboard others from.

to win a developer over, you have to win them over every single day.

Subscribe now

today’s drops 🔎

• apply to the $50k basis fellowship for operators, engineers, anyone

• cybersecurity internship @ crowe

• 1 month left of this $685k hackathon

the city skyline and skyscrapers are incredible. a sight to behold indeed, but there isn’t a feeling of rest. it’s unrest - the pull to be one of the last lights that shuts off and the bustle leaving my fingers itching to type something more.

perhaps that pull is quite necessary while i’m young and want to squeeze so much out of my hours. i don’t need the relaxation surrounding me all of the time - the hustle delivered to me through my environment may very well be the right call for my 20s.

regardless, i know i want to spend my life near a lot of trees.

i want to breathe deeply, go on a hike through rolling green hills whenever i want, and lie in a sunny patch of grass with a book in hand every morning.

there’s nothing as grounding as the literal earth.

no matter how much the city calls to me, i deeply cherish being surrounded by nature.

ideally, i get both.

Subscribe now

today’s drops 🔎

• check out this female founder dinner in nyc next month

• ai engineering internship @ postman

• applications just opened for the robertson full-ride scholarship

this bottleneck delays access to necessary care and hurts trust between patients and healthcare systems.

forus, founded in 2023, is an ai-powered network that connects doctors, pharmacies, payers, and biopharma to solve this problem. it’s a platform embedded within the physician workflow that automates insurance authorization, financial assistance, and routing for fulfillment for each drug and the involved stakeholders across america. more specifically, this means taking care of PA submission, appeal letters, benefit verifications, enrollment forms, buy-and-bill support, pharmacy routing, and patient status updates.

technically, it’s an EHR integration (which is why it’s easy for physicians to adopt). a doctor prescribes regularly, then agents generate the PA from the charts, submit, and monitor status. their verification layer is also interesting and worth a read here.

best of all, it’s completely free for providers and patients.

to date, forus has raised over $160 million. they’ve covered thousands of medical practices and health systems across all states, with over 80% of us zip codes represented.

questions i’d be interested to ask:

• biopharma pays, so what exactly are they buying? anonymized analytics, enrollment services, or influence over the routing?

• why won’t epic ship this?

Subscribe now

today’s drops 🔎

• zfellows $10,000 applications are open

• graphic design internship @ tarte cosmetics

• check out jpmorganchase’s code for good hackathon

i lived with the weight of my material possessions and the space they took up in my suitcase constantly in the back of my mind. i wouldn’t trade it for the world. in fact, once college is done, i’ll probably continue living like this well into my late twenties while i figure out life and what i want the rest of it to look like.

today i moved back into chapel hill for my junior year of college. i’ve moved into a house! alongside 3 amazing roommates, we’ve signed a 12-month lease. if all things go well, we’ll hold onto the place until we graduate as well.

my own room, my own space, for two whole years. no shifting around. all the time in the world to splay out frames and photobooth strips across my walls and fall into a pattern of physical familiarity.

perhaps i’m getting old, but i’m really excited to be physically rooted in one place for a bit.

i’ve been doomscrolling pinterest and mentally arranging my furniture for a whole month now, leading up to today. my entire family drove down from toronto to help me move in - my parents helped me carry so many things and build my furniture in the always-shocking chapel hill heat (35ºc today!). they are angels, and i wouldn’t be able to shape this 2-year home into what i’ve envisioned without their help.

i’m nowhere near done setting up, but i’m optimistic about falling in love with this house. you can sometimes see deer in the backyard, and a few stray cats that always come by in the evening. when i look outside my huge windows, it’s lush greenery. it’s a beautiful house, and filled with beautiful people.

i feel more excited about college and this upcoming school year than i did my freshman and sophomore years combined.

back to chapel hill. :)

Subscribe now

today’s drops 🔎

• apply for this $100k innovation challenge by august 16th here

• design engineering internship @ dedalus labs

• check out this hackathon happening tomorrow with a $1000 cash prize

it’s being intentional about following up and making plans; it’s putting yourself in the right rooms to find the right people; and it’s pushing yourself to push the boundary of what you know and experience with someone else. you certainly won’t make friends if you authentically sit alone at home.

the other moving piece that saves a lot of hardship and backtracking is having a good social filter. not placing biases onto someone upon first impression, but seriously considering if you’re thriving from a relationship or not.

does connecting bring you joy, laughter, growth, learning? anything positive without the need for secrificing your self? sussing that out quickly is the key to a well-oiled social filter.

the filter is for how this person affects your life overall. a friend in a rough patch who needs more than they can give right now is still a friend.

we meet plenty of people who suck the energy from us with nothing in return, hardly even a smile. perhaps you’re jaded by a few characters in your life that took advantage of your kind nature, or are simply a bad influence. perhaps they wave an ideal social life and a ticket to entry in your face, only to realize 2 years later, once you graduate, that you had nobody truly in your corner.

once you filter away the optics, proximity, and fronts, values and meaningful shared experiences tend to bubble up. we don’t need to click with every person we come across - and the more comfortable you are with that fact, the quicker it’ll feel like you find the right people.

there’s no need to dwell on the first people you meet in a new environment if they don’t make you feel good about yourself. it’s okay to wait for people who inject your life with love.

hold your filter to your values and trust in it.

Subscribe now

today’s drops 🔎

• apply to the $50k basis fellowship for operators, engineers, anyone

• product manager internship @ ibm

• 2 weeks left of this $180k google hackathon on ai agents