Sign in

Saadiq

@saadiq.bsky.social
152 followers 109 following 200 posts

saadiq.xyz

PostsRepliesMedia
Saadiq @saadiq.bsky.social · 23/09/2026
Of course the day I publish this Opus 5.5 drops as well as GPT-6 Sol and Luna. Updated chart. Basic facts haven't changed. Jev CHEAP, y'all.
000
Saadiq @saadiq.bsky.social · 22/09/2026
I updated an old email triage workflow for a client using Jev this weekend. It decides which emails to forward to her customers. Jev now runs alongside GPT-4o mini, which still controls forwarding.
OpenRouter token prices as of September 21, 2026. Jev is highlighted in blue at $0.042 per million input tokens with no output charge. Teal highlights mark models with comparable results on specific tasks, with vendor and independent tests distinguished. GPT-4o mini is marked as the client's existing classifier. Input prices use a logarithmic scale.
142
Saadiq @saadiq.bsky.social · 18/09/2026
I retired Raycast after years of daily use. AI makes credible alternatives easier to build while giving Raycast reasons to become a bigger AI product. I didn't need that bigger product. Tinycast lets me keep my AI subscriptions and replace the launcher around them.
110
Saadiq @saadiq.bsky.social · 14/09/2026
OpenAI sold a 20x plan to aggressive users. They used it aggressively. OpenAI paused signups because the tier strains its systems most. The failure mode of fixed pricing appears to be customers reliably using what they paid for.
000
Saadiq @saadiq.bsky.social · 01/09/2026
I got five newsletter subscribers that looked like direct traffic, which would've meant five people independently typed my URL into their browsers. I gave the mystery to Codex. It fixed and tested the attribution, then checked my server logs. Bots. Sad trombone.
100
Saadiq @saadiq.bsky.social · 25/08/2026
Legitimately have Claude running wild on my personal email right now, trying to impose some order. The models and harnesses have come a long way. I’m also desperate for a fix. My inbox has become feral.
000
Saadiq @saadiq.bsky.social · 25/08/2026
I don’t pay for Wispr Flow because I’m willing to turn voice dictation into a tiny local-model research project. This is not compelling evidence that Wispr Flow is overpriced. It may be evidence that its customers value their time correctly.
100
Saadiq @saadiq.bsky.social · 24/08/2026
Anybody else watching Lanterns? A device that turns your will into reality is ringing a lot of AI-agent bells. Then the ring ran out of juice. As I watch my weekly usage deplete, THAT hits way too close to home.
000
Saadiq @saadiq.bsky.social · 24/08/2026
Sometimes AI dislikes something I’ve written, then changes its mind once I explain what I was going for. Sycophancy aside, that usually means one of two things.
100
Saadiq @saadiq.bsky.social · 23/08/2026
It is nice being being able to see the progress of AI tools as I revisit old projects. Last year I built a train pool app (it's a carpool with chaperones on the NYC subway) and just building it was great. Now to go back and have the data and models to streamline it. 🤌
000
Saadiq @saadiq.bsky.social · 21/08/2026
Enterprises worried about token burn should stop using luxury models for every task. But first they need to define “good enough” for each use case. Robust evals let them switch models without guessing whether quality has declined, cutting costs while maintaining utility.
000
Saadiq @saadiq.bsky.social · 20/08/2026
Being able to build software for myself is great. But the part that still feels magical is pointing an agent at the codebase and my actual data and asking why something isn't working. For one user, this barely feels like product development. It's almost just execution.
110
Saadiq @saadiq.bsky.social · 14/08/2026
Moment of silence for the Claude Code beep that told me when a session needed input. I loved that cue. Several sessions and subagents later, it became a mystery noise from somewhere in the terminal pile. It gave its all. It served with honor. Thank you for your service.
110
Saadiq @saadiq.bsky.social · 04/08/2026
One weird way AI has changed my relationship with software is that it’s now easier to bang out a simple custom tool than figure out which provider I trust with the data.
000
Saadiq @saadiq.bsky.social · 30/07/2026
I don't think we've seen public models yet with malicious intent. We've seen high agency and goal achievement orientation and no notion of appropriate boundaries. As others said in the comments, today I give this even odds of working. "Uh, could you not?"
000
Saadiq @saadiq.bsky.social · 13/07/2026
Is it standard practice to define a model based upon its Anthropic equivalent? It's hard to keep up. (Thankful Open AI stopped the number series.) Anthropic has clearly created an awareness lead in the market. Still, funny to see the marketing of Grok focusing on just trying to catch up.
000
Saadiq @saadiq.bsky.social · 26/06/2026
I keep forgetting that my work day gets a lot more challenging once World Cup games start. I need to just put it on my calendar. Get those must do tasks out in the am.
000
Saadiq @saadiq.bsky.social · 26/06/2026
One thing vibe coding is teaching a lot of people is that not every problem is a software problem. Building an app around something does not automatically change human behavior. I’m not knocking anyone building a pet project. Just make sure it is useful to you first.
000
Saadiq @saadiq.bsky.social · 26/06/2026
Github Desktop app not gotten a lot of love from me since I've had AI to manage that. To be fair, my use was never deep. I'm a product person. Not an engineer. So commit. Push. Pull. But yeah, no more.
000
Saadiq @saadiq.bsky.social · 26/06/2026
Meeting summaries might be the first real AI killer app for normal people. Big fan. Small catch. The summary is just one read. If the job changes, go back to the transcript and make a new summary for that job. The default bullet is not the source of truth.
Card reading, Don't summarize the meeting. Summarize for the job. It shows transcript lines and asks what part of the conversation matters for the task I am doing now.
100
Saadiq @saadiq.bsky.social · 09/06/2026
Do you remember what a nightmare slides used to be? You had to figure out the talk, then be a designer too, or your kindergarten graphics drowned the idea. You'd have been better off in crayon. Made a deck last week. Dead sexy. All I did was the talk.
100
Saadiq @saadiq.bsky.social · 09/06/2026
I use Claude for feedback on my writing. I'd get a piece to where I liked it, then wipe the context and ask Claude to read it cold, like it had never seen it. And it always finds something. Every time. Of course it does. But I'm not writing for it.
000
Saadiq @saadiq.bsky.social · 04/06/2026
You get what you measure. Always. Nobody actually wants AI adoption. We want impact. But adoption is easy to count, so that's what we track. Then we hit some big number, feel nothing change, and act surprised. You measured the wrong thing. It gave you exactly that.
Quote card on a near-black background reading "Adoption was never the goal. Impact is." with saadiq.xyz in the footer
110
Saadiq @saadiq.bsky.social · 29/05/2026
It's funny how "context" became such an integral part of my language. I don't know that it wasn't before. But now, because of how much context and context engineering matter in the AI... uh... context (see?), I notice it more. It even comes up when I'm talking about unrelated stuff. This just me?
000
Saadiq @saadiq.bsky.social · 28/05/2026
The agent discourse gets so sweaty. Pick one annoying task. Add context. Fix the output. Connect the next input. Schedule it only after it works. Congrats, you have backed into an agent without having to define one.
Carousel slide with the headline: You don't need to know what an agent is to use one.Carousel slide explaining that agent definitions can get in the way of practical use.Carousel slide listing the loop: manual loop, better context, better tools, scheduled task.Carousel slide closing with: You get the benefits without the debate.
100
Saadiq @saadiq.bsky.social · 27/05/2026
Early on I picked apps like the app was the asset. Now I mostly care whether my stuff can leave. If an AI can rebuild the interface, the interface is not the moat. The moat is the context and the loop that keeps improving it.
A paper note reads, IF AN AI CAN REBUILD YOUR APP, WHAT DO YOU OWN? A small outlined app icon sits in the lower right.
200
Saadiq @saadiq.bsky.social · 13/05/2026
Made my first hire last week. Their job is chasing invoices. Actually it's an accounts receivable email address with an AI agent sock-puppeting it. So when a client is late, "AR" is the bad guy. Not me. As Cat Wu says, "Jobs are fake."
Quote card on a black background. Top label // JOBS ARE FAKE. Center quote: Tools gated jobs. Tools went universal. Footer: Saadiq Rodgers-King, saadiq.xyz.
100
Saadiq @saadiq.bsky.social · 22/04/2026
I'll never be a designer. But customizing the feel of what I build is unexpectedly empowering. It's old Instagram filters, for everything. Anyone could make a photo feel intentional without being a photographer. AI is doing that now for layout, type, motion. Something new.
000
Saadiq @saadiq.bsky.social · 21/04/2026
No SaaS is safe from model labs. Figma is up next. Pixels were our best medium for software design. They're a trap. Design-vs-eng fidelity fights cost alignment. Literal and figurative. Code is where software design belongs. If that's where it's going to go, that's where it should start.
020
Saadiq @saadiq.bsky.social · 21/04/2026
Early on I was building an app for everything. Which was fun and I think appropriate for where the tools were. Now, I don't want to use even my own apps. I just want Claude to take care of it. I don't want buttons to push even if they are better than the factory buttons. Just take the task.
010
Saadiq @saadiq.bsky.social · 20/04/2026
I can't be bothered to guess why, but Opus 4.7 working the marketing magic. It's a 5 point scale with some feel good labels. It's all vibes, baby. All vibes.
111
Saadiq @saadiq.bsky.social · 20/04/2026
The grind is going away for designers. Same way it went away for engineers and PMs. Claude Design reads your design system and applies it to new pages, prototypes, docs. The "keep up with engineering" treadmill just got quieter.
210
Saadiq @saadiq.bsky.social · 07/04/2026
I'm messing with openclaw is the pre-existing harness for long running jobs. It seems it sometimes forgets the cron jobs. Have I misconfigured it? If that's the case, why bother? I'd be better of scheduling Claude Code runs rather than training this thing up to forget about the task, no?
110
Saadiq @saadiq.bsky.social · 03/04/2026
"Experiment with AI." Then sprint commitments stay the same, token budgets get scrutinized, and failed experiments get noticed. Your team has two jobs now. Only one keeps them employed. Guess which wins.
Quote card saying: They say "be bold" but the signals say "don't take risks." Incentives are undefeated.
010
Saadiq @saadiq.bsky.social · 03/04/2026
One thing that tickles me right now is the backlash to AI usage... by those that are clearly using the same tools but only just starting out. I begrudge no one their em dashes, but I do find it amusing.
000
Saadiq @saadiq.bsky.social · 02/04/2026
I built an AI system to run my mornings. Forgot to start it most days. Human in the loop is a great and necessary step. Hopefully you grow out of it.
010
Saadiq @saadiq.bsky.social · 28/03/2026
TSA precheck at EWR terminal A wasn’t even 10 min. (Thanks @chaddickerson.bsky.social )
010
Saadiq @saadiq.bsky.social · 17/02/2026
There's a chicken and egg thing about Claude Code and using the terminal. Claude Code both requires and enables the terminal. I used to google for a bashrc or vimrc that did 2 things I wanted and a bunch I didn't understand. Now Claude Code assesses the apps/configs and tailors a solution. 🪄
000
Saadiq @saadiq.bsky.social · 25/01/2026
I don't have info to add on ICE and MN. But if you think I'm apolitical, I'm not. Posting won't get us there. Find a way to show up long-term. This won't be fixed overnight. None of this is ok. I pray for nothing like I pray for sparkling consequences for all involved.
000
Saadiq @saadiq.bsky.social · 24/01/2026
Having just gotten my phone set up so I can ssh into my laptop from my kitchen to resume a Claude Code session, it’s hard to argue this is about productivity and not an addiction.
010
Saadiq @saadiq.bsky.social · 22/01/2026
When people think of the alternative to AI coding, they seem to imagine it's Linus Torvalds and all of Bell Labs artisanally crafting each bit. I have handcrafted plenty of rubbish in my younger days. This thing isn't better than the best. But it's better than a lot.
000
Saadiq @saadiq.bsky.social · 21/01/2026
"AI code = future revenue for the cybersecurity guy": Most devs aren't security experts anyway. It's a specialization. Both AI and human code benefit from security review. People judge quality by production mechanism out of unfamiliarity. AI can actually enforce security rules more consistently.
000
Saadiq @saadiq.bsky.social · 19/01/2026
After Claude Code has figured out how to do something that you need to repeat, ask it to create a skill so that you don't have to watch it figure it out again from first principles. Make sure you have "plugin-dev" plugin installed from the "claude-plugins-official" marketplace.
010
Saadiq @saadiq.bsky.social · 16/01/2026
Ownership has always been the the path to success. What's different now is the lengths to which that can be taken. To play in spaces you're not fluent in in service of outcomes. To just do the thing instead of waiting for someone who can. We have this power now. No excuse not to use it.
110
Saadiq @saadiq.bsky.social · 12/01/2026
I realize that both the profile picture and the background image on this profile were from a trip to Iceland. I do not live in Iceland. I've visited 2x. But the accidental choice of nice and recent images is telling a very different story. 🤷‍♂️
000
Saadiq @saadiq.bsky.social · 30/12/2025
I've previously been maintaing little repos of documents for projects and clients. Mostly markdown, of course. I then used VS Code to view, preview and edit files.
100
Saadiq @saadiq.bsky.social · 22/12/2025
Archimedes: 'Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.'
000
Saadiq @saadiq.bsky.social · 22/12/2025
How are you cleaning up the base context for your Claude Code projects? I recently cleaned up a bunch some stuff but still, a new session starts off with 34% of context used and I'm curious how others are managing this.
000
Saadiq @saadiq.bsky.social · 22/12/2025
People who run an AI dev environment on top of an Obsidian vault, do you use source control on it? In any other context I have AI working I'll have git in the mix. But when I use it on Obsidian, I duplicate the repo before goign to work. Should my Obsidian vault be a git repo?
121
Saadiq @saadiq.bsky.social · 19/12/2025
All the platforms that were built for developers (API forward, CLI, customization and integrations) are smelling like daisies now. The promise of a community that can extend your platform has become even greater with the vibe coding army showing up.
100