Kol Tregaskes

DeepSeek v4 is coming, we just don't know when, not even they do.

x.com/i/status/20...

There is a real split happening between people who use AI as autocomplete and people who use it as leverage.

Autocomplete makes you a bit faster.

Leverage means you can define the task, direct the agent, validate the output, and run multiple workflows at once.

Same tools.
Very different outcomes.

AI Doesn’t Reduce Work—It Intensifies It One of the promises of AI is that it can reduce workloads so employees can focus more on higher-value and more engaging tasks. But according to new research, AI tools don’t reduce work, they consistently intensify it: In the study, employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so. That may sound like a win, but it’s not quite so simple. These changes can be unsustainable, leading to workload creep, cognitive fatigue, burnout, and weakened decision-making. The productivity surge enjoyed at the beginning can give way to lower quality work, turnover, and other problems. To correct for this, companies need to adopt an “AI practice,” or a set of norms and standards around AI use that can include intentional pauses, sequencing work, and adding more human grounding.

Source:
hbr.org/2026/02/ai-...

The research recommends organisations build intentional AI practices with pauses for assessment, sequencing to limit fragmentation and human grounding to preserve recovery and creativity.

- Employees took on responsibilities previously held by others, prompted tools during breaks and meetings, and ran multiple threads in parallel.
- A self-reinforcing cycle raised speed expectations and created workload creep.

This is my experience too. Generative AI intensified work rather than reducing it according to Berkeley Haas researchers.

- An eight-month observational study at a US technology company with about 200 employees revealed voluntary AI use led to a faster pace, broader task scope and extended hours.

Gemma 4:
x.com/AiBattle_/s...

Big week for Gemini, probably including Gemma 4 release.

x.com/AiBattle_/s...

- Shown on models up to 4B parameters trained on 200M images, 6M videos and 2M audio-video pairs.

This foundational work supports scalable multi-modal visual intelligence and world models.

x.com/i/status/20...

- Achieves up to 2.8x faster convergence versus standard flow matching baselines on key metrics (FID, FVD, FAD).
- Delivers better temporal consistency in videos, sharper text/typography, and joint video-audio outputs.

Black Forest Labs introduces Self-Flow - self-supervised flow matching for multi-modal generative models.

- Enables end-to-end training across image, video, audio and text without external representation models.

Zombies Per Minute on Steam Zombies Per Minute is a factory-building roguelite where everything you build is measured by one thing: how many zombies you can kill per minute. Yes, really. Use your imagination to grow your factory, make decisive choices each wave and progress from one outpost to the other. Until Earth is saved.

Wishlist it on Steam:
store.steampowered.com/app/4416970...

Try it:
www.zombiesperminute.com/

- Steam Deck verified with almost no extra work

Interesting tidbits:
- Entire game built 100% agentically with Codex + Claude handling all code, music, sound effects and voice
- Orbital AI companion CLAIRE guides the player Operator in real time
- Factorio-like 3D factory automation roguelite running smoothly in any browser

- 24 Feb: Declared project 100% AI engineered after 900 commits and 150 hours over one month alongside day job at Doctolib (€120 total AI spend)
- 6 Mar: GPT-5.4 via Codex fully revamped in-game tutorial with natural steps, smooth animated arrows, live demos, clear objectives and contextual hints

- 23 Feb: Shipped Attraction Tower, Copy/Paste, Final Stand and major late-game FPS boosts in just 3 days
- 24 Feb: GPT-5.3-Codex generated realistic zombie models and animations; reached 100k+ entities at locked 120 fps

Timeline to date:
- Late Jan 2026: Browser optimisation targeting 100k+ entities at constant 120 fps using React, Vite, TypeScript and React Three Fibre (no traditional engine)
- 11–14 Feb: Ran flawlessly on Steam Deck at 60 fps with zero changes; Claude Opus 4.6 used for main menu and UI work

Someone built a 3D Factorio-like game entirely coded with AI using Codex and Claude Code.

Playable on a web browser now, coming to Steam.
x.com/NicolasZu/s...

I would like, however, the other agents inside the messengers - that would be cool.

I’ll come back to OpenClaw in a few months.

PS. The /new and /reset does not fix the context overflow error I get a lot, and I'm using a 1M context model too!
x.com/levelsio/st...

Claude Code and Codex just work, and they have a nicer UI - particularly Cowork. So I’m looking to build my own multi-agent system, or use an off-the-shelf one instead, so I can get the 24/7 element that the CLIs cannot. Any recommendations?

I was working on a 2.0 version of my workspace, but the 24H2 Windows update borked my Windows mini PC (thanks Microsoft), so I’ve not had the time to rebuild it and carry on. Will do when I have the willpower.

I’m sure in the second half of this year, autonomous AI agents will "arrive" for general use.

It’s too early for this for me, but if you have it working then excellent. I’m not a coder, though, nor do I have the time to constantly fix something every day when it breaks. This is definitely for coders first, for now. It’s at its 2025 stage of AI agents.

Instead, it sent task updates and internal thinking, and only occasionally the news post.

It seems to reset at 4am every day too, and it completely wipes its memory of everything it was working on before. I tried channels and group topics, but it couldn’t figure out I just wanted results posted (e.g., news).

You have to regularly review what it does, because it does things like add more and more scheduled jobs when it doesn’t need to, or when it could be done more efficiently.

Fallback models fail. It keeps adding files that will never be read again, and it fills up its context (in its system files) to the point I get a context overflow error that can’t be fixed with a reset and having Codex/Kimi CLI resolve it.

It’s way too buggy: scheduled jobs fail a lot; it keeps forgetting (particularly how to use its own browser - every bloody day!!); it doesn’t read agents md; it always posts internal thoughts even after I tell it not to 100 times; and you have to manually update it.

I agree - OpenClaw is fun, and I like the convo element on Telegram, but it doesn’t work yet.