Pinned Tweet
I run an IT agency with AI agents. The thing that ate our time: writing SOPs by hand.
So I built stepkeep. Click through a task once, get a step-by-step guide with annotated screenshots.
No account. No upload. No subscription.
stepkeep.app
So I built stepkeep. Click through a task once, get a step-by-step guide with annotated screenshots.
No account. No upload. No subscription.
stepkeep.app
10
0
38
3,799
client's system rejected my e-invoice over a missing VAT id. fixed it, resent it, and now it's a different rule
2
0
4
95
in our agency we were leaving money on the table. Eyeballing it, we spent 75% of POs. Now we are oscilating around 100%. Not because we work more, but simply because we know more.
2
0
3
72
Google and Accenture launched Gemini agents for government workflows. Small teams just want a plan they can pay for by card.
1
0
5
108
switching models mid project means reloading the whole codebase context into the new model every time. what worked better for us: each agent works on its own git worktree and branch, and another agent merges and resolves the conflicts.
5
1
8
159
Nobody ready privacy policy (well, maybe germans lol) and the like only lawyers.
EU came up with GDPR not to protect consumers
but to protect corporations' businesses.
The company with bigger legal budget wins, so smaller competition is easier to kill.
EU came up with GDPR not to protect consumers
but to protect corporations' businesses.
The company with bigger legal budget wins, so smaller competition is easier to kill.
3
0
5
74
lol what can i say. EUrope is so good at teaching others what we dont know ourselves.
German chancellor Merz asks Jensen Huang for advice on how Europe should regulate AI.
Jensen: “You can’t regulate if there’s no technology.”
“The technology industry must invent the technology for safety. […] Safety is not hopes and dreams. Safety is technology and e companies need to have best practices. […] And then lastly there’s industrial and government regulations. […] Regulating the internet is hard, but regulating internet applications is not so hard. We should be regulating the applications right way. […] There’s nothing to regulate if the industry doesn’t thrive. You want the German AI industry to thrive as well as do it safely, responsibly. […] We know there’s technology we can invent and deploy in order to have AI safety. We know that there are best practices that we can adhere to as companies and industries. And then in addition to that, if necessary, we’ll make sure we regulate.”
1
0
4
59
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/De…
1/4
2
0
7
140
inception's mercury 3 doing 1,068 tok/s is only a win for agent loops where generation is the wait. i'd bet most pipelines lose more time on tool calls and review than on tokens, so the speedup shows up in a few steps, not the whole run. measure which steps are token-bound first.
4
0
9
116
I blink and there is a new break through in AI
1k tks is something i cant even imagine
1k tks is something i cant even imagine
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
2
0
5
132
Made a 60 s ad for stepkeep with AI agents in 1 hour. ~$12 in API calls.
- @GeminiApp for the cast
- @MiniMax_AI for the footage
- @ElevenLabs for the voice, sounds and music
- @HyperFrames_ to put it together
- @claudeai running all of it
How 🧵
- @GeminiApp for the cast
- @MiniMax_AI for the footage
- @ElevenLabs for the voice, sounds and music
- @HyperFrames_ to put it together
- @claudeai running all of it
How 🧵
2
0
4
146
What went wrong, and the fix:
- Faces. I prototyped on MiniMax H3-Max at 480p. Eyes went glassy, teeth smeared. Turns out H3-Max is the speed model and H3 is the quality one. Same keyframes on H3 at 2K: clean faces, ~$7.50 for the whole ad.
- Veo in Gemini caps at ~3 videos a day. I hit it on clip 2, so Gemini kept the stills and MiniMax took the video.
- Gemini gets confused as a chat grows. A fresh chat per batch fixed that.
- One take rolled the invoice up instead of unrolling it. Played it backwards.
- The stamps slammed down before the actors' gestures and hid the joke. Retimed each one to land after.
- The first mix was quiet on X (−21 LUFS). Normalized to −14 with a limiter.
- Faces. I prototyped on MiniMax H3-Max at 480p. Eyes went glassy, teeth smeared. Turns out H3-Max is the speed model and H3 is the quality one. Same keyframes on H3 at 2K: clean faces, ~$7.50 for the whole ad.
- Veo in Gemini caps at ~3 videos a day. I hit it on clip 2, so Gemini kept the stills and MiniMax took the video.
- Gemini gets confused as a chat grows. A fresh chat per batch fixed that.
- One take rolled the invoice up instead of unrolling it. Played it backwards.
- The stamps slammed down before the actors' gestures and hid the joke. Retimed each one to land after.
- The first mix was quiet on X (−21 LUFS). Normalized to −14 with a limiter.
1
0
2
218
Swapping the British narrator for an American one took a few minutes and cost cents. A traditional shoot of this would start around €6k.
stepkeep.app
stepkeep.app
0
0
1
14