| Jan 04 |
The best way to learn about agentic coding is to install one and painfully prompt your way into your first project.
Pain reducers: give it access to your full (backed-up) codebase and have it do a security scan. Or have it generate a few tests.
Just get used to this tech. Youāll have to sooner or later.
|
13 |
0 |
7 |
832 |
306 |
. |
| Jan 04 |
Ever since I started coding with agents instead of typing every line, coding has been fun again.
Particularly as a solopreneur, getting into deep flow is costly and often zaps energy from other tasks.
Watching Claude Code do that work and diving into the final PR is bliss. š§āš³š
|
69 |
4 |
25 |
3.7k |
280 |
. |
| Jan 04 |
And just to be clear, Iām talking about LLMs.
Iāve heard this countless of times: āI was always afraid to ask questions because veteran <xyz> would yell at me for being an idiot. And then, I discovered ChatGPT.ā
24/7 deep any-topic guidance without an ego.
|
201 |
1 |
8 |
7.6k |
264 |
. |
| Jan 04 |
This is what happens when you develop a technology that answers all user questions without judgement instead of laughing at them or calling their thoughts off-topic.
|
8k |
285 |
189 |
384.2k |
165 |
. |
| Jan 03 |
I, too, would love to be a spontaneous foreign policy expert.
Oddly, looking at my feeds, the people from back in my PoliSci university days, the ones I took foreign policy courses with, are not the ones tweeting ridiculous opinions.
Isnāt that something.
|
17 |
0 |
1 |
2.7k |
258 |
. |
| Jan 03 |
Don't just say "fix it. make no mistakes."
Instead, give it a process and milestones. Context + status-quo + reflection + resolution.
|
15 |
0 |
2 |
2.3k |
134 |
. |
| Jan 03 |
If you run Claude Code on a non-trivial codebase and use it for bugfixes, here's what works best for me:
- If a customer reported the bug, paste the report/convo verbatim
- Tell Claude Code to reproduce the error state before fixing it
- Then, explain it
- Then, have it fix it
|
188 |
3 |
34 |
12.4k |
278 |
. |
| Jan 03 |
Getting feedback from my readers is SUCH a joy! š„° https://t.co/gTNN8Mbngn
|
15 |
0 |
3 |
2k |
73 |
. |
| Jan 03 |
We have gone far beyond vibe coding.
If agentic systems are getting good enough for experts like Gergely to drastically change their toolchain, we're heading into a world where there is no question of "if" anymore.
|
520 |
27 |
27 |
71.8k |
215 |
. |
| Jan 03 |
Know-your-customer in one image.
|
21 |
0 |
7 |
3.8k |
32 |
. |
| Jan 02 |
I am realizing that my episode about the Dead Internet theory (and the onslaught of too many machines talking to machines) has the episode number of the 429 HTTP status code.
429: Too many Requests
Couldn't be a better fit if I tried.
|
20 |
0 |
5 |
3.9k |
236 |
. |
| Jan 02 |
It's not looking too good: https://t.co/minX4BINez
|
5 |
0 |
1 |
3.5k |
50 |
. |
| Jan 02 |
And if you're a software founder building or using AI tools and making it easier to integrate AI into people's workflows, are we making the internet a less human place? Can we make it more human, even though we embrace this technology?
|
2 |
0 |
3 |
470 |
235 |
. |
| Jan 02 |
The #1 streamer on Twitch is an AI.
Grok is undressing people all over the place, and the bots in the comments wage war.
The internet is just machines talking to machines. This is the core thesis of the Dead Internet theory.
So, am I contributing?
https://t.co/snxDFJ09H7
|
18 |
0 |
9 |
7.5k |
276 |
. |
| Jan 02 |
"Everyone knows it's a bubble" is the biggest bubble.
|
28 |
0 |
18 |
2.9k |
53 |
. |
| Jan 02 |
RT @bcherny: I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.ā¦
|
0 |
5.5k |
0 |
0 |
139 |
. |
| Jan 01 |
Tweets with this image always go viral š«¤
|
8 |
0 |
5 |
2.6k |
40 |
. |
| Jan 01 |
Weāre now in Q2 of this century.
|
17 |
1 |
5 |
1.6k |
32 |
. |
| Jan 01 |
https://t.co/BIWjUibyXI
|
4 |
0 |
1 |
2.8k |
23 |
. |
| Jan 01 |
I find the wave of grown adults asking Grok to create non-consensual "variations" of (mostly) women's photos in public quite disturbing.
It seems to be some people's first interaction with the concept of consent. It's equally strange to see the AI system still creating these images in the wild.
Can't believe that this is just the beginning. These tools willget much better, faster, and even less constrained over time. It already takes just seconds per image. In the past, the act of photoshopping this took a real person hours. Now it's a single inference pass. That's quite a difference.
Anyone who followed the artists-vs-AI conversation likely understands that this is likely a watershed moment for the public perception of AI systems when it comes to ownership of likeness, public distribution of these "generative results", and who is facilitating this at scale.
I give it one week for the EU to regulate this into oblivion.
|
21 |
0 |
8 |
8k |
936 |
. |
| Jan 01 |
Pretty big feature.
|
34 |
0 |
6 |
6.6k |
19 |
. |
| Jan 01 |
AI is the ultimate gatekeeper-breaker.
Don't understand a complicated error message? AI will explain it for free.
Don't know what that corporate email really means? How to say "no" courteously? Understand those rental contract clauses? AI helps.
Some people REALLY hate this.
|
64 |
6 |
33 |
4.7k |
278 |
. |
| Dec 31 |
Solid damage control.
|
2 |
0 |
0 |
1.3k |
21 |
. |
| Dec 31 |
haha, yeah, right https://t.co/cvNbMHAGDF
|
4 |
0 |
4 |
2.4k |
41 |
. |
| Dec 31 |
My wish for 2026 is for Claude Code to turn
"āŗ Now let me examine the XYZ model and related data structures:
āŗ Let me also examine the navigation structure and routes:"
into
"āŗ Successfully loaded my full understanding of your codebase from the 315 prior Claude runs."
|
37 |
0 |
6 |
2.9k |
274 |
. |
| Dec 31 |
My year in review: pretty amazing.
|
31 |
0 |
10 |
2.5k |
34 |
. |
| Dec 30 |
RT @__roycohen: @ganyicz @arvidkahl Two types of people, those optimizing for code generated by LLMs and people that think they're still goā¦
|
0 |
2 |
0 |
0 |
140 |
. |
| Dec 30 |
Of course, you can't rely on agentic code systems to write meaningful tests completely automatically. They've got much better at it, but you still need to read those tests to understand if the thing being tested is actually the thing you want to be tested.
But beyond that, access to testing has become so easy that now feels like I would just automatically do it on every new project because it is so easy. Just this tiny little improvement has significantly increased the quality of the software that I produce.
|
9 |
0 |
1 |
1.4k |
514 |
. |
| Dec 30 |
Before LLMs, I didn't test. I "had the full picture" (until a bug broke prod, of course.)
Because of LLMs, I learned how to test. I had it write tests for me, then fixed them to make sense.
You think LLMs write bad code? You should have seen mine before LLMs.
Much better now. https://t.co/TRGzAk2mTj
|
64 |
6 |
17 |
4.8k |
303 |
. |
| Dec 30 |
RT @marcelpociot: I'm super proud of our Expose growth.
In 2025 you shared ~500.000 sites with Expose - and around 80.000.00 (yes, 80 MILLā¦
|
0 |
6 |
0 |
0 |
140 |
. |
| Dec 30 |
RT @DsMatie: Then Zero to Sold by @arvidkahl
I wish this book existed when I started.
It explains how to build, grow, and sell a SaaS stepā¦
|
0 |
1 |
0 |
0 |
140 |
. |
| Dec 30 |
And here I thought hallucinating was an AI problemā¦
Good to see that gaslighting is still very much a human skill.
|
13 |
2 |
3 |
2.7k |
115 |
. |
| Dec 29 |
AI agentic coding splits developers into two camps: those who believe software must be correct, and those who think it will always consist of overlapping compromises.
And they battle it out on social media, assuming that they're talking about the same thing. "AI is great at this." "No, it's just hype." "Did you even try." "Didn't work for me."
But a no-side-effect single-purpose C library isn't just a different beast that a business-problem-solving web application, it's a completely different lifeform.
Arguing that engineering principles should be applied equally to both is not just short-sighted; it's impractical. Every single step from ideation to implementation has wildly different requirements.
So next time you see someone dismissing AI coding tools as useless, ask what they're building. And when someone swears it's a 10x productivity boost, ask the same question. The answer will tell you everything.
If you're building a cryptographic library, you need provable correctness. If you're building a dashboard that shows podcast mentions to a PR agency, you need it working by Tuesday. Both are valid. Both require skill. But only one of them benefits enormously from an AI that can scaffold CRUD operations faster than you can type the model names.
We're having a context collapse here, and people are too entrenched to realize they're talking about completely different things.
|
197 |
22 |
63 |
15.3k |
1.4k |
. |
| Dec 29 |
It's getting harder and harder to spot clear signs of AI-generated content.
A significant change is afoot.
|
33 |
2 |
12 |
5.3k |
107 |
. |
| Dec 29 |
RT @Yuchenj_UW: Claude Code was a side project at Anthropic.
ChatGPT was a side project at OpenAI.
PyTorch was a side project at Meta.
Gmaiā¦
|
0 |
631 |
0 |
0 |
140 |
. |
| Dec 28 |
Just reading through the install/usage docs of this excites me.
It's amazing to see the tooling surrounding an already spectacular toolchain. These efforts of fighting against lossy compaction are quite a step forward.
Can't wait to see this built into future version of cc.
|
37 |
1 |
2 |
9.8k |
276 |
. |
| Dec 28 |
We'll see a shift that's significantly bigger than the "mobile-first" responsive design wave or the SPA page boom, but this time, under the hood.
Things like expecting TOON or MCP support will become framework-level defaults.
And we'll design for human and agentic users alike.
|
34 |
2 |
13 |
6.5k |
279 |
. |
| Dec 28 |
https://t.co/yLXkbrITVi
|
36 |
2 |
6 |
6.9k |
23 |
. |
| Dec 28 |
Nerd drama is the best.
|
5 |
0 |
2 |
2.9k |
23 |
. |
| Dec 27 |
I am so close to a really running a āRewrite this Elixir project into functional Laravel. Make no mistakes.ā
And I have a feeling that it might actually work.
|
41 |
0 |
13 |
7.2k |
159 |
. |
| Dec 27 |
Embrace agentic coding or don't. Up to you.
But don't complain about "not seeing results" or "not getting it" without putting in the effort to learn a new skill.
Some people spend 2 years trying to fry the perfect egg. You might want to spend more than 15min with Claude Code.
|
312 |
14 |
51 |
18.2k |
278 |
. |
| Dec 27 |
RT @DaveShapi: https://t.co/bJM6zEcS56
|
0 |
186 |
0 |
0 |
38 |
. |
| Dec 27 |
RT @adamwathan: Is anyone using Claude Code or similar for stuff that isn't coding at all, purely because it's nice to have access to localā¦
|
0 |
23 |
0 |
0 |
140 |
. |
| Dec 27 |
This is why you MUST try to figure out agentic coding now.
If you canāt use it for your for-profit work, create a side project that is majorly built using agents.
Get a feeling for the loop, the models, and proper prompt technique.
Itās either that or becoming obsolete.
|
1.8k |
103 |
100 |
143.7k |
273 |
. |
| Dec 27 |
RT @claudeai: Starting at midnight PT tonight, all Pro and Max plans have 2x their usual usage limits through New Year's Eve. https://t.co/ā¦
|
0 |
608 |
0 |
0 |
140 |
. |
| Dec 26 |
I'm a big fan of Rob Pike and his work.
Why again is he raging against a thank-you email from software founders that was constructed and sent using a software tool?
Also, to deepen the subtweet: is this how we talk to each other now on social media?
|
23 |
0 |
16 |
10.4k |
251 |
. |
| Dec 26 |
Also, Europe: https://t.co/eXTfCB9mWF
|
27 |
1 |
6 |
2.7k |
37 |
. |
| Dec 26 |
As someone who was born in East Germany, where there was ONE kind of car, ONE kind of pocket calculator, really ONE kind of anything, this rubs me the wrong way.
The GDR had to come up with nonsensical slogans like "overtake without catching up" to account for a reality in which it was behind in technological and social advancement.
Now, long after that system crumbled, I find myself as a technologist, and the pride in having locked yourself into a single technical choice (with significant hurdles to innovation) sounds hollow to me.
|
34 |
1 |
8 |
7.4k |
540 |
. |
| Dec 26 |
The fine https://t.co/9riNyvYZc5 has this wonderful visualization of how Claude's Max pricing at $200 is absolutely worth it. https://t.co/OcWzDYDciE
|
83 |
1 |
11 |
10.6k |
149 |
. |
| Dec 26 |
Cloudinary is worth $2bil and is effectively ImageMagick-as-a-service. LaunchDarkly? If-statements.
These services are never just "literally" doing anything. It's the "aaS" in SaaS that makes these things so valuable.
They figured out all the little hard things along the way.
|
92 |
5 |
18 |
16.9k |
278 |
. |
| Dec 26 |
And if you have any disdain for vibe-coders, ask yourself why you don't mind people being beginners and amateurs in other fields.
This is an embrace-or-be-left-behind situation.
|
96 |
1 |
5 |
11.6k |
178 |
. |
| Dec 26 |
Claude Code is a significant step towards allowing the person who has the challenge to also implement the solution.
I say this as a developer, seeing most of my profession being compressed into a toolchain.
I know I'll always be useful. I know what good code and architecture look like. That alone is a craft worth honing.
But seeing non-technical people build things that genuinely help them without being gouged by dev agencies is wonderful.
|
1.1k |
54 |
71 |
405.7k |
446 |
. |
| Dec 26 |
RT @rahulgs: thank you claude + playwright mcp for doing my fsa claims this year https://t.co/9L6byYJWHF
|
0 |
61 |
0 |
0 |
104 |
. |
| Dec 26 |
Please feel free to subscribe to the newsletter at https://t.co/c4fJ5q9Hno. Once a week, deep insights into a founder-relevant topic.
|
0 |
0 |
1 |
1.3k |
133 |
. |
| Dec 26 |
You'll find the podcast episode for this at https://t.co/snxDFJ09H7
|
1 |
0 |
2 |
1.3k |
67 |
. |
| Dec 26 |
Everyone is celebrating Christmas, so who keeps engaging? š
Today's TBF newsletter issue is all about the "Dead Internet" theory and how we, as SaaS founders and technologists, can avoid contributing to a world where only machines talk to machines.
https://t.co/nkj6cQVN8o https://t.co/L122c4oqKc
|
6 |
1 |
8 |
2.3k |
298 |
. |
| Dec 26 |
If this kind of prompt looks stupid to you, consider that this is the simplified version of āin all places of where we need to change the padding to make it look uniform, change p-4 to p-8ā and all of a sudden, it becomes a powerful tool.
|
34 |
0 |
19 |
6.3k |
238 |
. |
| Dec 25 |
Assuming an image is AI will be the new assuming an image is not AI.
|
11 |
0 |
9 |
2.6k |
68 |
. |
| Dec 25 |
So many things boil down to taste and idiosyncratic preferences.
In a large enough market, almost anything will find a customer.
|
37 |
0 |
8 |
6.7k |
130 |
. |
| Dec 25 |
RT @alexalbert__: PSA we're doubling Pro/Max plan usage limits from midnight tonight through NYE.
Go build something cool :) Happy holidayā¦
|
0 |
146 |
0 |
0 |
140 |
. |
| Dec 24 |
True sales enablement.
|
10 |
0 |
3 |
3.6k |
22 |
. |
| Dec 24 |
Merry Christmas š„°
|
44 |
0 |
16 |
2.3k |
17 |
. |
| Dec 23 |
If you run production loads on these models, youāll understand. Prompts needed to get things working reliably take a lot of work, and theyāre not easily migrated between models.
Model versions are dependency risks, and they expire quickly.
|
45 |
0 |
7 |
7.9k |
240 |
. |
| Dec 22 |
Hear me out: I don't want every single service I use to do a "2025 wrapped", showing me just how much they monitor every single one of my in-app activities and how they compare to my lookalike customers.
|
30 |
0 |
8 |
2.3k |
203 |
. |
| Dec 22 |
So glad I could contribute to their business success by being in the 27th percentile for product usage! š„°
|
13 |
0 |
3 |
1.8k |
105 |
. |
| Dec 22 |
All I want For Christmas is product-market-fit.
|
189 |
11 |
41 |
8.7k |
47 |
. |
| Dec 22 |
The article about the fire is here: https://t.co/Wrj1BoTr5O
My fav song is this one :) https://t.co/eIOxKEPENC
|
2 |
0 |
0 |
1.5k |
111 |
. |
| Dec 22 |
Bus factors don't just exist for bootstrappers.
VOLA, an (objectively great prog) metal band, had their touring equipment ($200k+) go up in flames when a storage facility fire broke out. Insurance will pay $30k (lol), everything is gone.
So far, fans have pledged $70k in their GoFundMe.
Looks like they'll be able to recover from their fans alone. I've seen some fans donate thousands of dollars. True fans, you could call them, in Kevin Kelly's vernacular.
Between merch sales, tickets, and these donations, they'll find their way back into profitability. The lesson for me as a founder is that disaster strikes whether we like it or not, but recovery depends on a strong presence with the people who care about your work. Engaged supporters MATTER.
Also, you should really listen to VOLA.š¤
|
11 |
0 |
5 |
2k |
798 |
. |
| Dec 22 |
Haha. the race is on :D
|
23 |
1 |
6 |
4.2k |
23 |
. |
| Dec 22 |
If you're not using your SaaS's blog to teach your ideal user how to pitch a budget for your tool to the purchaser within their company, you're missing out on one of the most impactful ways of doing content marketing. Just saying.
|
100 |
2 |
27 |
6.9k |
230 |
. |
| Dec 22 |
So, what are people already doing?
I've seen people monitor their competitors' changelogs. Have agents run on a daily or weekly basis. Scrape the changelog compared to what they know in the past, summarizing new features, pricing changes, even terms and conditions changes, and reporting that back to you. So you always keep tabs on every single one of your competitors and what changes in their business for you to react to or maybe anticipate.
People have been building documentation auto-updaters that scan outdated documentation in the background. Definitely, and this is what I use it for is inbound lead researcher when new trials sign up, I figure out who they are and what they might be interested in to get some customized personal outreach.
For SEO content gap finder, it's really cool, particularly if you know who your competitors are, so you can see what they might be missing. Then, if you're into refunds or dispute evidence data collection, you can have an agent go through somebody's database traces and compile a full report to file with their bank if they should choose to file a chargeback.
And then there's just the never-ending task of mining for customer success and failure stories on social media, forums, websites, and all over. Anything you can do to stay on top of these conversations and categorize them and look for opportunities is something you can hand with a well-defined prompt to an agent that can just constantly keep manually tracking these things.
You could quite literally tell your Claude code to check the front page of 20 different subreddits or Facebook groups one after the other with a 5-minute wait interval in between and see if something new has been highlighted that could be useful for you to focus your attention on. This could be something an agent is doing 24/7 for weeks and will always highlight very interesting things for you to deal with. You might even automate the outreach at that point too using a different agent. It is incredible what you can do.
|
149 |
8 |
4 |
10.9k |
2k |
. |
| Dec 22 |
With Claude Code, it is now possible to have a permanent at-the-very-least-intern-level assistant doing 24/7 research jobs for you and also building the software tooling to act on that data. Admin panels, automated media conversion, building ideal libraries, and monitoring competitors. All that and more.
Between MCPs, Cloud for Chrome, and Skills, if you don't have your agentic tool churn through your work backlog all day long, you are quite literally missing out.
Because every day, more and more of your competitors will do exactly this. And while a lot of this currently is clunky, haphazard, and DIY, you can be absolutely sure that this will be commoditized within the next six months.
|
1.7k |
100 |
55 |
105.4k |
697 |
. |
| Dec 21 |
Claude in the browser will break many "API access is for enterprise customers only" services.
I tasked it earlier with finding some data in a popular SEO analytics tool and presenting it to me in JSON format.
Sure, took forever, but it's now just another automated workflow.
So now, when you consider that any browser session might be an agent controlling the browser chrome while a human is watching (or likely doing some other thing), your UI is just yet another API.
And it's not really scraping, either. It's a lossy, currently screenshot-based, super-slow version of what it will be in a few years.
But it's here today, and it's HIGHLY automatable. People WILL use it to grab more data than they manually could. Parallel sessions. Headless automation beyond Puppeteer. All controlled by agents that never stop.
You might really want to start rate-limiting ANY frontend route. Because the zombies are coming.
|
369 |
23 |
39 |
32.7k |
918 |
. |
| Dec 21 |
Tried Google's AI Mode for the first time, and it immediately suggested @levelsio's favorite topics. š¤£ https://t.co/uw2FOZJ2iy
|
28 |
0 |
13 |
5.5k |
126 |
. |
| Dec 21 |
Deleting all incoming emails is technically inbox zero.
|
58 |
0 |
17 |
10.3k |
55 |
. |
| Dec 21 |
I wonder when someone's going to release `left-pad` as an agentic "skill."
|
10 |
0 |
1 |
2.2k |
74 |
. |
| Dec 21 |
RT @flaviocopes: TailwindSQL is a game changer https://t.co/l0e1jhyWR3
|
0 |
42 |
0 |
0 |
70 |
. |
| Dec 20 |
Any field of expertise eventually reveals this truth: it is not stirring the sauce that is complicated and takes years of experience; it is knowing the right ingredients, pots, spoons, temperatures, and precise steps that make it hard.
Same goes for coding, writing, public speaking.
The true value lies in an expert's ability to deploy their taste to judge quality and to discern good work from bad work at a glance.
|
12 |
0 |
2 |
4.3k |
419 |
. |
| Dec 20 |
So are we still doing MCPs? Or is it skills now?
Been thinking of making Podscan more "in-AI-usable." Right now, it's a REST API with all the trimmings, but I wonder how I can make it more pluggable.
|
18 |
1 |
9 |
4.8k |
200 |
. |
| Dec 20 |
RT @terminaldotshop: https://t.co/O7emEOf88U
|
0 |
5 |
0 |
0 |
44 |
. |
| Dec 20 |
RT @GergelyOrosz: Amusing how everyone (who is not a dev) who assumed *writing code* is the hard part about sw development cannot comprehenā¦
|
0 |
109 |
0 |
0 |
140 |
. |
| Dec 19 |
Machines changing their point of view when presented with evidence is my favorite thing. Particularly because people have such a hard time doing that.
āStrong opinions weakly heldā is the LLM way.
|
13 |
0 |
3 |
2.6k |
197 |
. |
| Dec 19 |
I dream of a permanent, exploratory background code-review agent.
Just constantly scans for opportunities to improve, fix issues, prevent bugs, strengthen tests, all the time, pre-/mid-/post-commit.
It'll play havoc with our understanding of a persistent codebase though. Hm.
|
78 |
1 |
30 |
8.3k |
277 |
. |
| Dec 19 |
This is a repeat of what happened when computers and digital case lookups were introduced to the legal industry.
Anyone who preferred flipping through tomes in libraries was quickly left behind.
Back then, lookup and retrieval were revolutionized.
Now, it's reasoning.
|
8 |
1 |
1 |
2.4k |
271 |
. |
| Dec 19 |
The moment we'll run out of Science Fiction stories to tell, I'll start believing in the Singularity.
|
10 |
0 |
6 |
1.5k |
101 |
. |
| Dec 19 |
RT @StoyanBuilds: @arvidkahl Likewise, I just listened to the whole episode and jotted down the essentials. Thank you so much for sharing tā¦
|
0 |
1 |
0 |
0 |
140 |
. |
| Dec 19 |
Today on the pod, I share how I, a dev who really isn't into marketing, have found a way to still get it done.
(Turns out, you can delegate a lot of it to machines, you just have to know what, and how. And what NOT to delegate.)
https://t.co/IMy7QU2Qyt
|
60 |
1 |
8 |
6.7k |
254 |
. |
| Dec 19 |
And THIS is why we need generative AI :D
|
6 |
0 |
1 |
2.6k |
40 |
. |
| Dec 19 |
This is the kind of stuff you can expect enterprise companies to already be tinkering with in their software engineering labs, trying to find a way away from the AI platforms hosted and controlled by somebody else, particularly in industries where regulation prevents most cloud software usage.
This is the kind of work that will facilitate adoption of AI tooling into these companies, and it becomes critical for software companies to consider that not only do they have to do on-premise deployment of their software stack, they also have to be able to integrate into the on-premise deployments of their AI clusters.
|
3 |
0 |
2 |
838 |
619 |
. |
| Dec 19 |
4 Mac Studios connected via Exo using the new RDMA thingy.
Consumer hardware (albeit still expensive), sizable LLMs, 1.5TB of VRAM. 30+ tok/s on DeepSeek 3.1 (the 671B model).
Bit expensive to vibecode, but completely private, permanent, customizable.
https://t.co/aF7zJnuBuA
|
8 |
2 |
3 |
2.9k |
278 |
. |
| Dec 19 |
Kinda weird that every other person in my parentsā generation warned me not to trust anything I saw online when I was a kid in the late 90s.
The same people now consider this to be an amazing and totally real thing, without a flicker of a doubt. https://t.co/qiBKuoEk5V
|
8 |
0 |
6 |
2k |
270 |
. |
| Dec 18 |
https://t.co/WRkrXjLAdc
|
7 |
0 |
2 |
2k |
23 |
. |
| Dec 18 |
Remember that the ONE thing your churned customers will remember is how easy it was to part amicably.
If you want to fuck this up, I recommend:
- call-to-cancel
- no-refund policy
- dark patterns in every interaction
- vendor lock in
(Serious note: Olly is doing things right)
|
13 |
0 |
3 |
2.1k |
280 |
. |
| Dec 18 |
Sometimes I watch the map of people currently using Podscan, and I'm mesmerized by the fact that this project is being used by professionals all over the world. https://t.co/HvoyIH7E4y
|
31 |
0 |
3 |
1.9k |
184 |
. |
| Dec 18 |
Allowing non-technical people to create AI slop has been the most impactful form of technology democratization since self-publishing enabled everyone to share their precious thoughts.
|
26 |
2 |
6 |
1.6k |
183 |
. |
| Dec 18 |
RT @manic_pixie_agi: https://t.co/scmvfcgHSv
|
0 |
77 |
0 |
0 |
44 |
. |
| Dec 18 |
Is there already such a thing as an "external hardware LLM" like we have external hard drives?
Instead of having to run/maintain a local model, I want an inference machine that I can just plug in and point my prompts at. Single GPU, maybe a few in parallel.
Who's building this? https://t.co/snA0TH8PN6
|
3.1k |
151 |
637 |
402.7k |
304 |
. |
| Dec 18 |
In 2026, weāll see a mainstream shift in virtual agentic employees.
Paying for an āautonomous marketing agentā with full access to your blog, email, and socials will turn from weird to useful.
Crafty founders already build and run these things internally. And they convert.
|
14 |
0 |
8 |
1.9k |
275 |
. |
| Dec 18 |
RT @TheMechFrog: If we all have more children, they can be put to work on these wheels to generate power for the AI data centers.
|
0 |
17.9k |
0 |
0 |
129 |
. |
| Dec 18 |
RT @pingthebird: This is real btw. It did everything by itself https://t.co/eWZWVQ22hb
|
0 |
119 |
0 |
0 |
86 |
. |
