Michael Myburgh (@MPMZA_) header
Michael Myburgh (@MPMZA_) profile picture

Professional AI agent babysitter. Employed meat proxy. Paid to say “no not like that”. Emotional support human for bots

Cape Town, South Africa
Joined July 2026
Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)Photo by Michael Myburgh (@MPMZA_)
I built /cockpit for Claude Code. It's a sidebar that shows what the main thread and every subagent are doing, live Once Claude launches a few agents in the background, the chat tells me they started and then I am stuck waiting and I wanted something different, I wanted to see which one is still working and on what, which one has gone quiet, which one is stuck until I approve a command and what the batch has cost me so far So every agent gets a block with its elapsed time, tool calls, errors, context size, estimated cost, model, and the tool call it's on this second. An agent with no activity for 3 minutes turns yellow, and one waiting on approval turns red. Click an agent to read its transcript, stop it, or send it a note The main thread has its own block with a context bar and a countdown on the prompt cache. Background shells, monitors and workflows are listed too, and I get a notification when the batch finishes, so I can go work on something else github.com/MPMyburgh/claude-…
3
2
6
429
Michael Myburgh retweeted
With all due respect to all the builders, When the game demos going to end? Don't get me wrong I like them but stopped because almost all models are pretty good now and we're passed that, 3d is solved. Nobody is making a full game actually and just blasting demos, commit to an idea and complete it because then it's worth sharing and trust me you'll feel much better after making something actually real for people to use.
56
5
2
98
5,147
Maybe one day we will stop benchmarking new models based on blender, threejs and html page oneshots, none of these matter for real world performance 🫠
1
3
46
Google is going to force Anthropic to drop either Fable 5.5 or Opus 5.6 next week, calling it now 😂 IPO hype was there, sharp drop with release of Carbon and Neon now they need to take the podium back
74
Looking to connect with some techies, those battling AI psychosis all day, running agents 24/7 hit me up with the coolest thing your agent did this week 👇
8
4
231
Being 4 hours away from home and just being able to ping my homelab agent to check things is pretty cool, everyone should have this
2
102
Michael Myburgh retweeted
who should I follow
2
1
1
7
264
News and normies rn
1
2
56
@carthurj we mogging the normies today?
1
1
33
I am having an absolute ball working with all the new Gemini models, Hydrogen, Carbon & Neon, they are fantastic, glad there are more hats in the arena again, it's just healthier for end users, competition forces changes that benefit us non-enterprises ultimately!
54
I have been running Fable 5.2 primarily since the release and making use of my resets to keep my usage limit up to date, applied it in my workflows, audited my codebases and skills, this is my read: - There is some regression regarding orchestration, it is FANTASTIC at long horizon tasks but it is extremely eager to read and review and that burns the context windows unnecessarily - It is really good for exploring complex topics or finding creative ways forward when you're blocked - It is really good at architectural decisions - It is subpar at coding but extremely good at writing agent briefs (which is why it sucks that it needs guidance for effective orchestration) - Opus 5.5 is still the GOAT, just using /advisor fable is literally enough for just about anything, when you're stuck on a tough task just ask Opus to check-in with the advisor on the way forward No reason to burn limits on it for software imo
2
1
53
Gemini Neon is absolutely nuts, Google are back baby, 3 rapid fire releases and all of them have been bangers, if this doesn’t scream RSI I don’t know what does, so glad we’re back at the stage where all the frontier labs are fighting for the podium again!
62
Just to be clear, when you put rules in your global AGENTS.md or CLAUDE.md or even repo level, you need to know that these can and will be ignored, by making use of permissions/hooks etc you need to block and redirect agents mechanically, putting "do not delete my home drive" doesn't mean anything
3
40
Blessed by the king of LLM's himself
Replying to @MPMZA_
so dario wrote the bad reviews
1
21
Making good use of the the new themes in Claude Code
20
Looking forward to the OpenAI drop today, going to be lit
11
On the same machine and the same model, a fresh Claude Code session starts at 20.9k tokens in the CLI and 49.3k in the GUI. Most of the gap is the MCP. The GUI loads 16.6k of tool definitions upfront, while the CLI tools loads its tools on demand and starts at 0, and the GUI's system prompt is 11.1k against the 3.5k For one session that's fine, since 49.3k is 5% of a 1M Window, but I lean very heavily on subagents, and each one starts its own context and pays that overhead again, so your usage takes a massive hit I am very much still married to the CLI, the GUI has some really nice QOL features but that's about it
3
2
65
It’s concerning that everyone is critiquing Qwen4-max for only somewhat improving over Qwen3.8-max and blaming it on Western labs blocking distillation. Even if that’s the case, the slowdown and death of open source LLMs should concern us more. A monopoly would look a lot different from what we have now with Anthropic and OpenAI: non-existent limits and crazy pricing. The Chinese models put a lot of pressure on the pricing and service of the Western labs. DeepSeek Flash v4.1 led to price cuts across all the major labs. Just keep it in mind!
23
Just got uncalled for access to a new flash model… Had to divert my eyes and call the police, wouldn’t recommend
1
35
My usage reset is on a Sunday evening, I make an active effort to burn through 100% by Friday, AI makes it too easy to feel productive and for dopamine starved brains it’s addictive, at least it allows me to switch off over weekends, anyone else?
1
25
Insane that we’re living in our version of the dot com boom and a lot of people are complaining about usage limits instead of watching AI go brrr
16