- https://xitter.cf/t.co/S9Rs0fPO77 - Claude code for content creators - https://xitter.cf/t.co/gHWQNDD0Ep - https://xitter.cf/t.co/KqM3b6w9et - https://xitter.cf/t.co/3iqKOs76rG
Joined May 2008
- Tweets2K
- Following196
- Followers232
- Likes240
hashin retweeted
opus 5.5 can make launch videos you can sell for $1/2k
my first opus 5.5 animated video took me 30 minutes
this one took 4
i built a skill for my agentic coding setup that automates most of the process
-> vc-backed startups regularly spend $3/5k on launch videos like these
-> and with every vibe coder shipping a new product, demand has never been higher
rt + comment “skill” and i’ll send you the full workflow
(follow so i can dm you)
Was testing out Astra on a 5x plan
Was going to upgrade it to 200
Turns out it is pretty good
But $200 plan is paused
5x with Astra does not get shit done
Why openai why?
Bought a muse.ai subscription for £10
Basically use that for without browser use.
RIP rate limits!
I've been able to maximise the rate limits with this My MAX 200 plan
Compact often but without it forgetting.
Desktop app.
say "continue this work in a separate session OR - spawn a new task using spawn_task"
A chip opens and waits for you
/compact
change the chips run condition to "run in this session"
Click on it ! Boom
Narrow beats broad with AI agents right now.
The agents that actually work today are scoped to a single job:
- Write a post
- Transcribe a meeting
- Answer one category of support questions
The "general purpose agent that runs your business" is further away than the hype suggests.
Zuckerberg said it himself.
Cold DMs get ignored. Comments under competitor posts don't.
Someone asking "anyone know a tool for X" in a comment section is a lead raising their hand in public.
The move: find the comment, reference exactly what they said, offer something specific, then ask.
Response rates run 3-5x higher than cold outreach. Context beats a script every time.
I stopped chasing every new AI model.
Instead I broke my work into six jobs: writing, building, images, video, editing, and automation. One default tool for each.
The job picks the tool, not the launch hype.
A new video model drops every few weeks, and chasing each one is how a week of real work quietly disappears.
When something trends, I ask one thing: is my current tool actually failing this job? If not, I keep working.
Meta stock had its best week since early 2024, up about 15%, on optimism about its AI strategy.
The market isn't rewarding announcements anymore. It's rewarding shipped product people actually use.
That's the bar now for anyone building with AI too. Ship something people use, not something that just sounds impressive in a keynote.
Meta launched its first AI image model, Muse Image, built by Meta Superintelligence Labs.
Users are already pushing back because it can use public Instagram photos to generate new AI images.
If you post publicly, "public" now doubles as training data by default. Worth knowing before you post your next photo, not after.
Illinois just signed its AI Safety Measures Act into law.
Another state, another set of rules for anyone building or selling AI tools.
If you're a solo builder shipping AI features, compliance isn't optional anymore just because you're small. Worth a read before you scale into a state you haven't checked.
I stopped comparing my AI stack to everyone else's on X.
Someone's always running six tools I've never heard of, glued together in ways I'll never see the seams of.
My stack is the one I actually finish things with. That's the only comparison that matters.
Meta is reportedly close to a deal to run Anthropic's Claude model on its own private AI infrastructure, per SemiAnalysis.
Two years pushing its own open models hard, and it might end up running a competitor's model under its own roof.
Says more about how hard frontier AI is to build alone than any keynote will.
Sundar Pichai just said it plainly: Google is losing the agentic coding race to Anthropic and OpenAI.
More compute, more data, more cash than almost anyone alive, and the CEO still said the quiet part out loud.
Infrastructure doesn't automatically win. Whoever developers reach for does.
Meta says its updated model, Muse Spark 1.1, beats OpenAI and Anthropic on certain benchmarks.
Notice what's missing: no comparison to either company's most recent release.
Benchmark claims from any lab are marketing until you run your own task against them. I stopped trusting a chart the moment it started picking its own competition.
OpenAI quietly pulled its own recommendation for SWE-Bench Pro.
Their own audit found 30 percent of the 731 coding tasks in it were broken.
A benchmark that a lab retracts on its own is more trustworthy than one nobody ever checks. Still doesn't mean the number you saw quoted last month was real.
Everyone's talking about AI agents. Nobody's talking about which tool actually runs them.
Zapier: easiest, 6,000+ apps, from $29.99/mo. Built for non-technical teams.
Make: visual builder, stronger branching logic, from $10.59/mo.
n8n: open source, full code control, free self-hosted or $24/mo cloud. Built for developers.
Most people start on Zapier and graduate to n8n once they know exactly what they need.
Quiet fact that doesn't fit the AI-is-killing-jobs headlines: software developer job postings are ticking up since Claude Code launched, per Indeed's Hiring Lab.
Not exploding. Ticking up.
Worth remembering next time someone states with total confidence what AI is doing to the job market. Nobody actually knows yet.
OpenAI just launched ChatGPT Work to go head to head with Claude Cowork.
Two labs now building the same bet: an agent that handles real office work, not just chat replies.
When the biggest AI companies start copying each other's product shape this fast, that's the signal the category is real, not hype.