🌱 Liam

488,600

September 2026 points

Scores updated 7 min ago

Tweets: 121
Likes: 3,948
Retweets: 44
Replies: 130
Quote tweets: 20
2× boosted: 5
Followers: 703+43

Every scored tweet

300 per tweet · 100 per like · 200 per reply · 300 per retweet · 300 per quote tweet

Tweet	Likes	Retweets	Replies	Quotes	Boost	Points
thank you @NYC_DOT for fixing this! concrete barriers installed, problem solved ✅ https://t.co/RpGgpFNXqH https://t.co/ZXTntNOiID Aug 2🥉 Bronze	2,904	32	19	5	—	305,600
@Liface @NYC_DOT nope there’s a ton of improvements being made around here! lots of people pointing out long standing problems like this & they’re finally fixing some of them Aug 2	201	0	1	0	—	20,600
probably my favorite moment of the day is walking into the stillness of @fractaltechnyc in the morning and feeling the possibility of everything that can be created here https://t.co/xhwrNgWcBN Aug 5	24	1	3	0	2× Fractal	7,200 3,600 × 2
there's a guy in @fractaltechnyc writing C code...by hand Aug 4	21	0	3	2	2× Fractal	7,200 3,600 × 2
they’ve still got those old sodium orange lights around mccarren and I hope they never change them https://t.co/LASHIHMho5 Aug 1	37	0	4	1	—	5,100
live notes from fractal accelerator class Aug 1	13	1	2	0	2× Fractal	4,600 2,300 × 2
what do all of these things have in common? wrong answers only > Fractal, Maximum New York, Vibecamp, Cuties, Community Archive, Psychofauna, Whorehole, The Service Guild, Claude Squad, Casa Tilo, Solis, Concrete Kingdom, Vibechella, and ApplyAll Aug 5	15	0	1	0	2× Fractal2× MNY2× Vibecamp2× Cuties2× Comm. Archive2× Psychofauna2× Whorehole2× Service Guild2× Claude Squad2× Casa Tilo2× Solis2× Concrete Kingdom2× Vibechella2× ApplyAll	4,000 2,000 × 2
first time i’ve ever seen solar punk graffiti https://t.co/pK8LJCqwnR Aug 4	24	0	6	0	—	3,900
a harness that helps users manage their own attention is going to unblock so many people https://t.co/YM98Dp2mzs Aug 6	24	1	4	0	—	3,800
best door dash dropoff pic so far https://t.co/ZnvFmH5EUb Aug 1	30	0	2	0	—	3,700
if a spot like this floods every time it rains, does the city keep track of that anywhere? https://t.co/FVHhh67CXz Aug 17	21	0	6	0	—	3,600
[performative anti-AI man in Brooklyn]: AI Slop? I prefer Aēsop Aug 19	20	0	1	2	—	3,100
A New Yorker's guide to Manhattan https://t.co/pFJOQzPFpp https://t.co/LEFkEB4Isn Aug 5	19	0	2	1	—	2,900
I feel like fruit and music came from the same place Aug 2	17	1	2	0	—	2,700
https://t.co/03CNVS9jKd Aug 2	15	0	3	0	—	2,400
I think the most exciting opportunities for better agent UX come from unbundling the chat for example, having a record of decisions I've been asked to make, separate from any specific chat Aug 4	13	0	2	1	—	2,300
when editing UI with agents, its really nice to break everything into a component library and iterate on individual, composable UI components making a design system is so easy now! why not do it for every project Aug 19	12	0	2	1	—	2,200
@NYCMayor @NYC_DOT the new super-wide Ashland Place bike lane is amazing, but it’s already full of parked cars any plan for a physical treatment to block cars from entering? 🙏 https://t.co/GlAHGGwOyf Aug 16	17	0	1	0	—	2,200
chat management is under invested in look how excited people are for pinned projects / chats in codex, or scheduled chat creation think about how small the chat management feature set of most harnesses is relative to the in-thread feature set Aug 5	15	1	0	0	—	2,100
basically all of the problems I have with built-in connectors for chatgpt / claude are due to multi-tenancy I just need to be able to connect multiple gmail accounts, multiple calendars, multiple notion accounts, etc Aug 3	12	1	1	0	—	2,000
this feels like a bigger lesson for building software with agents slop comes from trying to "print" the whole thing at once vs. breaking down its components and building them one by one @__drewface demo'd a technique to me yesterday for low-poly video game asset creation that followed a similar philosophy Aug 19	14	0	1	0	—	1,900
there needs to be a "rails" but for artifacts that you manipulate commonly with agents, standard patterns of artifacts that everyone ends up reinventing locally (if this resonates and you think you know what these are I would love to hear your thoughts) Aug 19	8	0	4	0	—	1,900
let go and let claude Aug 15	12	0	2	0	—	1,900
realizing that a lot of using agents is just unlearning my learned helplessness around reshaping the computer to support the way I want to think & act Aug 10	12	0	2	0	—	1,900
instead of websites, we now have The Website Aug 19	11	0	2	0	—	1,800
the correct way to eat a bagel is top-down: eat the top slice of the bagel first, then the middle stuff, then the bottom slice of the bagel Aug 7	9	0	3	0	—	1,800
one of the reasons working with agents is hard is that it requires you to begin running an ad-hoc scheduler in your head, across multiple threads. computers do this too! and its pretty hard to get right, even for them. why would it be any easier for us? https://t.co/kuSWdmp10G Aug 4	10	0	1	1	—	1,800
Gardening is solved, pests are not yet solved. https://t.co/zVmQ1xRJDp Aug 23	12	0	1	0	—	1,700
"we think that it’s useful to contemplate the quadrillion-dollar world" https://t.co/clowEzClfZ Aug 19	14	0	0	0	—	1,700
plan mode https://t.co/r7S5RWWAdT Aug 5	10	0	2	0	—	1,700
[braindamaged from 10 years of Slack]: "what if the user experience of the future was an endlessly noisy chat room?" Aug 7	13	0	0	0	—	1,600
lmao https://t.co/j9Y0XXKnw3 Aug 5	13	0	0	0	—	1,600
what are your compounding daily loops? https://t.co/udMYz68PuS Aug 4	7	0	3	0	—	1,600
what loops in your life compound? Aug 4	4	0	3	1	—	1,600
@__drewface @kailajlim @fractaltechnyc COMPUTER CHURCH Aug 1	5	0	0	0	2× Fractal	1,600 800 × 2
the computer is an infinite video game Aug 31	7	0	1	1	—	1,500
restate in simple, plain, practical language, assuming I have zero context Aug 26	8	0	2	0	—	1,500
there’s a giant hawk living in a floodlight in mccarren park Aug 19	10	0	1	0	—	1,500
everyone should have a Personal Application Aug 10	9	0	0	1	—	1,500
read park bench plaques Aug 10	9	1	0	0	—	1,500
its pretty hard to make a feedback loop. most feedback never loops Aug 5	12	0	0	0	—	1,500
ideal agent interface https://t.co/cMS0CsCxTh https://t.co/Y6nqeJmiSa Aug 3	12	0	0	0	—	1,500
stories are the best memory systems send tweet Aug 23	8	0	0	1	—	1,400
https://t.co/AzDXajOfMA Aug 20	9	0	1	0	—	1,400
this is the best intro to context that i’ve ever seen https://t.co/4yfMJq6S8r Aug 14	11	0	0	0	—	1,400
i love the sun Aug 7	7	0	2	0	—	1,400
@Liface @NYC_DOT @LincolnRestler has been advocating for more fixes to streets near Atlantic & Flatbush Aug 2	11	0	0	0	—	1,400
The New Colossus 2: A Newer Colossus “we’re gonna need a bigger door” https://t.co/Z3PcFqryn6 Aug 1	8	1	0	0	—	1,400
you may not be interested in queues, but queues are interested in you Aug 4	7	1	0	0	—	1,300
debugging why an agent didn't work is usually like 5 things: - did you really tell the agent what outcome you wanted in sufficient detail? (prompts) - if the agent needs a detailed, step by step process to accomplish the outcome, did it discover and follow that process? (skills) - did you give the agent access to files / data that it needs? (context) - does the agent have access to everything it needs to take the action for you? (tools) - did the model just stop responding? did an outage happen? did the computer or process crash where the agent was running? did some integration not respond? (platform) Aug 1	7	1	0	0	—	1,300
Right — and that observation is the pattern paying rent. Aug 9	9	0	0	0	—	1,200
come to our weekly demo party next Wednesday! demo something cool :) https://t.co/PwvsJxfGxf Aug 7	3	1	0	1	—	1,200
the real eval was the friends we made along the way Aug 26	8	0	0	0	—	1,100
@venividinotavi I was just talking to my friend John Jacob Jingleheimer Schmidt about this Aug 18	6	0	1	0	—	1,100
✅ https://t.co/mOA34NUhua https://t.co/CjNiRYBFR8 Aug 7	8	0	0	0	—	1,100
what if your computer told you something surprising about yourself every day Aug 6	8	0	0	0	—	1,100
bring back the Walk speed limit https://t.co/p9qvOE7czA Aug 3	8	0	0	0	—	1,100
@__drewface pointed out that pi is a computational process that’s Out There running forever and we just sample it for equations, kind of like electrons coming out of a steam turbine Aug 2	6	0	1	0	—	1,100
happy birthday jerry garcia Aug 1	3	1	1	0	—	1,100
@__drewface also its easy to fall into a "gambling" loop with agents, where you "roll the dice" on outcomes over and over, and the variable-intermittent-reward loop of "will I get a good result" this time drives you deeper into a crash-in mania loop Aug 1	6	0	1	0	—	1,100
@__drewface when your cognition is augmented, you get popped up to other bottlenecks / frontiers of your own personal development, psychological, emotional, behavioral, environmental, cultural Aug 1	6	0	1	0	—	1,100
@wordrotator https://t.co/97cBswCLNb Aug 21	7	0	0	0	—	1,000
and yet https://t.co/scEHuOsZYM https://t.co/8dU9LYym33 Aug 15	7	0	0	0	—	1,000
https://t.co/GqooBH1npG Aug 12	7	0	0	0	—	1,000
@Prigoose @venividinotavi this is what tpot wars does to a man Aug 7	3	0	2	0	—	1,000
crashing out = nothing is possible crashing in = everything is possible crashing in is "ai psychosis / mania" - you can do it all, right now, forever @__drewface Aug 1	5	0	1	0	—	1,000
easy setup for agent “dreaming” put markdown notes in a github repo and run a nightly action that lets claude/codex tidy up & consolidate them Aug 10	4	0	1	0	—	900
@__drewface another hard part about agents is that you don't know whether the agent loop will run long or short and its difficult to manage a process where the timing of outcomes is constantly shifting, constantly unpredictable Aug 1	4	0	1	0	—	900
@__drewface reflections on "default ways of processing" in agent world being manic ADHD context switching 5 tabs, 40 agents, 4 tasks ...but this is something like the skillset that a CEO or operator of a complex system runs into mgmt of attention in complex systems is possible to build Aug 1	4	0	1	0	—	900
@__drewface the antidote to "crashing in" might be something like "satisficing" -- i.e. developing the skill of developing "good enough" solutions rather than optimizing https://t.co/BxhUpvv9aV Aug 1	4	0	1	0	—	900
@__drewface most agents pick some boundary of abstractions over data + chat + context, but they all overlap in weird ways and different ways and do not make it easy to gain fundamental mental models + skills for using this stuff AGI might be here, but UX is still bad Aug 1	4	0	1	0	—	900
@__drewface "now I understand why when I told my grandma, "just buy a computer" she felt overwhelmed" Aug 1	4	0	1	0	—	900
chain of thought is kind of like self narration i.e. telling a story about what you’re up to and what’s you’re doing next - but why does the story need to be told in the first person and only within a short time horizon? https://t.co/s8cLd06GgY Aug 23	2	0	0	1	—	800
@Prigoose wait who’s the xerox man?? Aug 20	3	0	1	0	—	800
@Prigoose @venividinotavi i’m experimenting with bait for points Aug 7	3	0	1	0	—	800
https://t.co/yG67T89BgH Aug 7	5	0	0	0	—	800
https://t.co/LpeV34O41W Aug 5	5	0	0	0	—	800
@Liface please write this Aug 1	3	0	1	0	—	800
@lorenz_system we need one empire state building sized statue of liberty with 1000 people living inside of it Aug 1	3	0	1	0	—	800
@__drewface on the other hand, there's no need to compromise when software frustrates you, you can build everything you want. also you can just draft multiple solutions in parallel, rather than pick one Aug 1	3	0	1	0	—	800
@__drewface "where's my stuff"? and "where are my projects?" are two really difficult problems to solve for most people getting started with agents why isn't this file from claude cowork available when I start the claude chat or when I'm in claude code or when I'm using chat gpt work Aug 1	3	0	1	0	—	800
@__drewface claude cowork / claude code / claude design / https://t.co/nN0JQdrB4m chat / chatgpt / chatgpt work / codex etc. on web, mobile, desktop the product surface of using agents is just massively overwhelming for consumers Aug 1	3	0	1	0	—	800
ampcode orbs are amazing Aug 16	2	0	1	0	—	700
@__drewface “decomplect, make no mistakes” /goal Aug 5	2	0	1	0	—	700
@puheenix lmao Aug 4	2	0	1	0	—	700
@puheenix and pats Aug 4	2	0	1	0	—	700
@aaaronwill bodurger Aug 31	3	0	0	0	—	600
the computer is an agency training environment https://t.co/JgCaRrI6LB Aug 31	3	0	0	0	—	600
are trees agents? Aug 23	1	0	1	0	—	600
@pee_zombie too late Aug 19	1	0	1	0	—	600
@pee_zombie CIA bird robots confirmed Aug 19	1	0	1	0	—	600
@thorstenball already using the test flight app every day! but it seems like a mobile-responsive web app wrapped in a native shell, rather than true native experience hoping someday to use orbs within an app that feels as good as the codex / cursor native mobile apps Aug 17	1	0	1	0	—	600
@wordrotator but what’s the best album Aug 5	1	0	1	0	—	600
@zencephalon this is amazing, I used to do this with index cards Aug 4	1	0	1	0	—	600
@zencephalon habit decks? pls explain Aug 4	1	0	1	0	—	600
@Liface @pav_light “I’m free all of the time” Aug 2	1	0	1	0	—	600
@aaaronwill lmao Aug 31	2	0	0	0	—	500
@wordrotator exactly Aug 27	2	0	0	0	—	500
@thorstenball native mobile app for amp code orbs would change my life 🙏 if it had codex level polish I think I would literally never use another coding harness Aug 17	0	0	1	0	—	500
@__drewface @just_be_dev https://t.co/PwvsJxfGxf Aug 7	2	0	0	0	—	500
@aaaronwill stores your air in a table Aug 4	2	0	0	0	—	500
https://t.co/iyVHfDUMoK Aug 2	2	0	0	0	—	500
@zencephalon @puheenix thank god Aug 2	2	0	0	0	—	500
bots give me "execution in the kingdom of nouns" vibes Aug 29	1	0	0	0	—	400
@puheenix it’s my /restate skill Aug 27	1	0	0	0	—	400
@aaaronwill LMAO Aug 23	1	0	0	0	—	400
@pav_light https://t.co/ByO0bxFJrX Aug 20	1	0	0	0	—	400
@pee_zombie lmao Aug 19	1	0	0	0	—	400
@pee_zombie lfg Aug 19	1	0	0	0	—	400
@ledflyd @NYCMayor @NYC_DOT https://t.co/OFanZS5MEY Aug 16	1	0	0	0	—	400
@__drewface so true Aug 8	1	0	0	0	—	400
@just_be_dev @__drewface https://t.co/PwvsJxfGxf Aug 7	1	0	0	0	—	400
@wordrotator lmao Aug 5	1	0	0	0	—	400
@conundrumer gottem Aug 5	1	0	0	0	—	400
@zencephalon now I want to know what’s in all of your decks! Aug 4	1	0	0	0	—	400
@choosy_mom big one Aug 4	1	0	0	0	—	400
@wordrotator no one can stop me Aug 1	1	0	0	0	—	400
agent loop = describe the plot ♻️ advance the plot https://t.co/Ici20N8Y7T Aug 23	0	0	0	0	—	300
* agents in the broadest sense of the word Aug 23	0	0	0	0	—	300
i.e. claude / codex default styles Aug 19	0	0	0	0	—	300
[braindamaged from 10 years of Slack]: "what if all of work was an endlessly noisy chat room?" Aug 7	0	0	0	0	—	300
Total · 121 tweets in September 2026	3,948	44	130	20	5 boosted	488,600

Posts, replies, and quote tweets count — native retweets are excluded. Mentioning a project on the boost list doubles that tweet's points.