Snapdeck — prompt a deck, export PPTX, no watermark. $10
fishkill, NY
Joined July 2025
- Tweets2K
- Following51
- Followers50
- Likes1.8K
manzi retweeted
Grok Bot maxxing for marketers and GTM
We put everything we and our friends know into this one
github.com/bigmacfive/ulpaso
if you try the transcript, tell me where it breaks.
🤖 Made with AI
manzi retweeted
Anyone using Codex/Copilot CLI
Do you guys have tricks you use to shrink the built-in system prompt?
Claude Code gives you some levers you can pull - disabling built-in tools and features.
Does Codex/Copilot have the same?
manzi retweeted
Codex GPT 5.6 over engineering tv show.
Tibo, any tips to stop GPT-5.6 Codex from over-engineering? @thsottiaux .
----
Me: Give me a lighter to light a cigarette.
Codex: No. You need a tank.
manzi retweeted
Which one do you prefer for coding?
- Cursor
- Copilot
- Antigravity
- Claude Code
- Codex
- Blackbox AI
manzi retweeted
codex plus users after reading this to the end
Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex. I had mentioned this a while ago, but then postponed it.
This is necessary as (a) the 5h limit allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience.
We are for the upcoming months keeping the 5h limit not enabled for Pro $100 and Pro $200 subscriptions.
This is the year of AI agent. I might not have posted it here last year, but I sure know it's the natural extension after GenAI. However, what I wasn't expecting is the amazing acceleration of harness from theory to real adoption. OpenClaw, Hermes Agent, Codex, Claude Code, Grok CLI, Google Antigravity, DeepSeek harness, Pi, and more.
The trend toward separating the model (the reasoning engine) from the harness (the runtime scaffolding) has changed how AI software is built. With the harness controls the execution loops, tool definitions, context management, and token efficiency, it's important to evaluate the right harness which often dictates the success rate and execution cost far more than changing the underlying model.
Big corporate will stay with closed model and harness, similar to how Windows OS or Mac is the safer route instead of running Linux. But with all these open model and harness, it will be a game changer as it will highly disruptive on even replacing the OS itself.
Think about it, most of the work can actually be done through a browser now. There is no reason to not have a highly customized industry/business related OS/AI that easily connects to all necessary departments. Turning any of these harness with agents and adding human in the loop will be the best streamlined workflow and ultrafast execution.
The question now is - how soon will this happen?
Sometimes you wake up at 2am and can't fall back asleep. Solution? Give your Claude and Codex agents a few tasks, get a leg workout in, then check your Discord for the results. Adding a study lane to my free exam study site, tweaking fam financials secure site, posting a journal about our recently built exam prep site's skill for adding said study lanes with one sentence. Day is going well so far :P
manzi retweeted
wdym codex has built-in interactive graph explainer! I usually just ask them html and then delete it later, but it nice to just have it in the app. It's too dense and hard to read in this state but hopefully our lord @thsottiaux will iterate more on it 🙏
manzi retweeted
After using Claude Cowork, Codex and Cursor for 1000's of hours, this morning I span up copilot "work" / cowork .. (who even knows what your actually using).
Background: I am doing a series that compares them so wanted to include Copilot.
I have 38 sample invoices which I wanted summarise.
Simple enough.
In Claude, Codex (ChatGPT), Grok, Cursor, I can just point it at the local folder and it just works. Brilliantly.
In the desktop Copilot app, I am first confronted with "Web" or "Work" ..
"Work" means it can "see" my work files.
So here's what I tried ..
1. Try to get it to see my local folder .. nope .. It can't access local files.
2. So I zipped them and and ... nope .. you can't upload a zips.
2. I then tried uploading the files and ... nope .. you can't upload more than 20 files
4. I tried putting them in OneDrive and ... nope .. It couldn't find the folder or the 38 files.
The task was a fail.
Let's just say, Microsoft have some serious catching up to make their tooling as easy as Claude / ChatGPT / Grok Bot have made it.
Microsoft have a massive opportunity, because they have such a wealth of a companies data already locked into their ecosystem.
The competition don't have the data..yet.
But, if MS dont sort out the naming / branding and ease of use, companies will be forced to use the simpler alternatives.
manzi retweeted
Built this mini Aim Lab within a day using @sparqworlds + Codex before I start working on a horror FPS
Still wild to me that I can build something like this with zero coding experience
SPARQ in 5–6 months is going to be crazy
manzi retweeted
Same OpenSpec apply + archive, same repo.
Codex · gpt-5.6-sol · high
weekly quota used: 10%
Claude · Opus 5 · extra-high
weekly quota used: 3%
5-hour quota used: 27%
manzi retweeted
wouldn't it be nice if Claude Code and Codex were just zsh "unknown command" fallbacks?
wouldn't need to --resume or switch foregrounds
It was great fun participating in @swyx's Kill My SaaS competition the other week, lots of impressive submissions.
As we wait on results, I thought it would be fun to get codex to spin up a video with remotion and elevenlabs demoing my submission.
## eval competition idea: Help kill my SaaS
my team is proposing to pay >$40k/year for enterprise saas we have never used and will never be able to customize.
as a smol business owner, this feels shitty.
thinking of doing a small remote hackathon:
- i cover $1000 in tokens for you
- you do your best to clone this SaaS in a weekend
- my team (your prospective customer) evals it
- winner gets $10,000 cash & @latentspacepod writeup
- all code is open sourced
everyone wins except high margin low moat saas.
we keep doing this with increasingly ambitious saas things for SMBs until we find the boundary of what saas is still hard to kill in a weekend.
does that work?