John Kennedy

3K posts

John Kennedy banner
John Kennedy

John Kennedy

@CommerceJohn

Founder @actual_ai_ . Previous Head of Product @AWSCloud GameTech. @UnderscoreVC Core Partner. Also 🦋https://t.co/T2K4x7mAt0.

Seattle, WA Entrou em Temmuz 2012
682 Seguindo1.2K Seguidores
geoff
geoff@GeoffreyHuntley·
what i’m coming to understand is software modularity is more important than ever before. as the agents are forgetful what ya gotta do is push stuff down as an infrastructure concern. let’s take logging/tracing - move that from an application level concern to an infrastructure middleware/effect concern that provides application tracing for free that way when the clanker forgets to sprinkle it in control flow it doesn’t matter. less the agent has to do the better outcomes thanks for the convos @xoofx i think you are right.
English
21
9
140
9.8K
Arel Avellino
Arel Avellino@ArelAvellino·
@CommerceJohn @garrytan @claudeai Been running Claude Code as an OS for a few months now. The knowledge codification piece is the part most people haven't figured out yet - the system gets smarter over time, not just faster.
English
1
0
1
61
John Kennedy
John Kennedy@CommerceJohn·
Further more to Claude actually being an operating system, skills can auto-update, checkout how @garrytan does it:
John Kennedy tweet media
English
0
0
3
237
John Kennedy
John Kennedy@CommerceJohn·
This is why @openclaw needs to be on @awscloud or @hetznercloud, not your desktop.
Andrej Karpathy@karpathy

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

English
0
0
0
107
JB
JB@jamie247·
Just spent $500 in compute vibe coding my own Civilisation RPG but with unbounded natural language diplomacy.. meet Uncivilised. ask me anything.
JB tweet media
English
225
120
2.9K
320.7K
John Kennedy
John Kennedy@CommerceJohn·
Meta-prompting is coming to every discipline, but prompts are stone tools. There are generalizable processes in most white collar work. Gather information, converse for clarity, generate, review, etc. In the coming months, better prompting will not win. Process engineering will take us to the next derivative.
Ihtesham Ali@ihtesham2005

A Stanford student got reported for academic misconduct last semester. His research paper was so good his professor assumed he bought it. The academic integrity hearing lasted 3 hours. Here's what happened in that room. The panel asked him to explain his methodology from scratch. He opened his laptop, pulled up Kimi.com, and started rebuilding the entire paper live in front of them. First he fed it his raw notes and asked: "You are a research methodology expert. Here are my raw notes. Identify the 3 strongest arguments buried in this data, rank them by originality, and show me exactly where each one challenges or extends existing literature." The professors went quiet. Then he ran: "Now simulate a hostile peer reviewer with a PhD in this field. Generate every serious objection they would raise against my thesis. Then tell me which objections actually have merit and which ones I can dismantle." One professor leaned forward and asked him to stop so she could write down the prompt. He kept going. "Take my weakest argument and steelman it harder than I did. Show me what it would look like if it were airtight. Then tell me what I'd need to prove to get it there." Then the one that ended the hearing. "You are my thesis advisor. I have 24 hours before submission. Read this draft and tell me the single change that would move this from a B+ to an A. Be brutal." He walked them through how he'd used that last output to rewrite his conclusion three times until it held up under every objection in the room. What took most PhD candidates 6 months of back-and-forth with advisors, he was doing in real-time inside a single workflow. The panel didn't just clear him. They gave him the highest grade in the department's history and asked him to present the workflow to faculty. The irony is beautiful. The paper looked too good to be human because he'd found a way to think harder than most humans bother to. That's not cheating. That's the new ceiling.

English
0
0
0
73
Damian Player
Damian Player@damianplayer·
local in-person AI classes for corporate boomers. think Claude 101. easily a $ 25K/mo opportunity. rent a presentation room. run meta ads targeting 35-60 year olds. charge $500-$1000 for a 2-day hands-on workshop. teach vibecoding, Claude, ChatGPT, prompting and agents. the demand is insane. these people see AI everywhere but have zero clue how to use it. they want face-to-face guidance, not online courses. run the same curriculum weekly. refine based on questions. multiple up-sell or down-sell opportunities. scale to multiple cities once you nail the format. you’re hitting a market everyone else ignores. corporate boomers with cash who prefer learning in person. they are also being told to learn these tools daily.. no chance this doesn’t work if you execute. go out and nail this.
English
146
81
2.9K
245.8K
Jo
Jo@JoJoFromJerz·
@Acyn My god — he just makes shit up as he goes along. We haven’t seen any evidence of any of this and he’s not ever going to show us any evidence of any of this because there isn’t any evidence of any of this. We’re in a war with Iran because Israel told us to. That’s it.
English
61
319
3.1K
27.3K
Acyn
Acyn@Acyn·
Doocy: You said: they hit Qatar, Saudi Arabia UAE, Bahrain, Kuwait nobody expected that. We were shocked. Are you surprised that nobody briefed you ahead of time that that might be their retaliation? Trump: Nobody. Nobody. No no no no. The greatest experts—nobody thought they were going hit…
English
1.7K
3.5K
12.3K
3.4M
Chris Murphy 🟧
Chris Murphy 🟧@ChrisMurphyCT·
It’s crystal clear now that Trump has lost control of this war. He badly misjudged Iran’s ability to retaliate. The region is on fire. 1/ I’m going to explain to you in this🧵what I’ve learned - in part from closed door briefings - about the four biggest current crises.
English
4.3K
10.1K
41.3K
5.5M
John Kennedy
John Kennedy@CommerceJohn·
In case anyone *still* needs to be disabused of the notion. Do not use New Relic. The usage-based charges are ridiculous.
English
1
0
2
129
John Kennedy
John Kennedy@CommerceJohn·
@Austen Which audio generator / model are they using?
English
1
0
1
1K
Austen Allred
Austen Allred@Austen·
Apparently the kids in my son’s school (Alpha School) have been using AI to generate songs to help them memorize their spelling words. He always makes his heavy metal. I have to admit this would have 100% worked on me.
English
39
25
613
82.6K
Austin King
Austin King@MadeWithOzten·
@CommerceJohn I dodged this ADR process (left 4 years ago), but you're giving me AWS quarterly process PTSD 🤣
English
1
0
1
12
John Kennedy
John Kennedy@CommerceJohn·
@addyosmani Finally can switch from gog! Every company is going back to CLI.
English
0
0
2
54
Addy Osmani
Addy Osmani@addyosmani·
Introducing the Google Workspace CLI: github.com/googleworkspac… - built for humans and agents. Google Drive, Gmail, Calendar, and every Workspace API. 40+ agent skills included.
English
654
1.6K
15K
5.4M
John Kennedy
John Kennedy@CommerceJohn·
Depreciation Criteria: What needs to be true to deprecate this functionality. (Critical in the world of software factories)
English
0
0
1
100
John Kennedy
John Kennedy@CommerceJohn·
@andrewchen This is fundamentally wrong. Agents will recommend skills/CLIs/SDK to other agents. Truly, if you build it (and it is good), they will come.
English
0
0
0
11
andrew chen
andrew chen@andrewchen·
AI is makes software cheaper to build, but not easier to distribute This is the core conflict for years to come
English
340
117
1.4K
148.3K