Daily Intelligence Briefing: Tech / AI / IT Monitor

Date: March 11, 2026

Executive Summary

OpenAI's GPT-5.4 model appears to be rolling out, with team members suggesting significant improvements particularly for mathematics and science learning. NVIDIA released Nemotron 3 Super, a 120B mixture-of-experts model with 12B active parameters and 1M context length, now available through Ollama with fully open weights. Perplexity announced "Personal Computer," an always-on local AI product, sparking debate about what truly constitutes "local" AI. Replit secured $400M at a $9B valuation while launching Replit Agent 4, and infrastructure reliability concerns emerged after an OAuth outage impacted AI research operations. The developer tools landscape continues evolving with Claude Code adding scheduled automation features and growing criticism of hosted CI provider reliability.

Key Events

• OpenAI GPT-5.4 appears to be launching with team members indicating the model performs particularly well for interactive math and science learning, though details remain sparse → link

• NVIDIA Nemotron 3 Super released on Ollama: 120B mixture-of-experts model with 12B active parameters, 1M context length, designed for multi-agent applications with fully open weights and datasets → link

• Replit raises $400M at $9B valuation while launching Replit Agent 4, marking significant momentum in AI-powered development tools → link

• Perplexity launches "Personal Computer" - an always-on local AI product that triggered debate about true local computing, with Framework responding that genuine "local" requires offline capability on owned hardware → link

• Infrastructure reliability concerns surface: Andrej Karpathy reports OAuth outage wiping out autoresearch labs, highlighting vulnerability of AI-dependent workflows to service disruptions → link

• Claude Code adds scheduled automation: Ollama announces ability to run prompts on schedules within Claude Code, enabling automated tasks and reminders → link

• GitHub Actions reliability issues drive migration: Developers reporting switching to Blacksmith after persistent GitHub Actions failures and OOM errors on ARM builds → link

• OpenAI shares technical lessons on agent computer access: Insights on making long-running workflows practical for AI agents → link

Analysis

Infrastructure Maturity Concerns: Multiple signals indicate growing pains in AI infrastructure—OAuth outages affecting research, GitHub Actions reliability issues, and CI provider problems. This suggests the industry is hitting scaling limits that require architectural rethinking rather than incremental improvements.

Local vs Cloud AI Debate Intensifies: Framework's pointed response to Perplexity's "Personal Computer" launch highlights a fundamental schism about what constitutes true local AI. As models grow more powerful, the definition of "personal computing" in an AI context becomes increasingly contested, with privacy and control implications.

Developer Tools Evolution: Karpathy's observation that "we're going to need a bigger IDE" where the basic unit is an agent rather than a file signals a paradigm shift. The launches of GPT-5.4 (for learning), Replit Agent 4, and Claude Code automation features all point toward development environments designed for human-AI collaboration at higher abstraction levels.

Open Source Momentum: NVIDIA's release of Nemotron 3 Super with fully open weights, datasets, and recipes represents continued commitment to open AI infrastructure, potentially countering concentration of power in closed models.

Watch Next: GPT-5.4 capability assessments as users gain access; infrastructure provider responses to reliability concerns; Replit Agent 4 adoption and competitive responses from other AI coding assistants; resolution of the "local AI" definitional debate as more products launch.

Tweet Feed

AI Model Releases & Updates

@ollama · 2026-03-11T17:02

NVIDIA Nemotron 3 Super is now available on Ollama.

ollama run nemotron-3-super:cloud

🦞Try it with OpenClaw:

ollama launch openclaw --model nemotron-3-super:cloud

Run it locally on your device:

ollama run nemotron-3-super

120B mixture of experts model with 12B active made for multi-agent applications 1M context length fully open weights, datasets, and recipes → tweet link

@gdb · 2026-03-11T15:36

team really cooked with gpt-5.4 → tweet link

@gdb · 2026-03-11T01:54

such suspense — gpt-5.4 pro (potentially) for open mathematics: → tweet link

@gdb · 2026-03-11T00:59

ChatGPT for interactive learning of math and science: → tweet link

@jezell · 2026-03-11T05:45

RT @sukh_saroy: Holy shit... Fudan University just built an AI that generates vector animations from text, images, and video.

It's called… → tweet link

@jezell · 2026-03-11T05:39

RT @ErickSky: 🚨Acabo de ver TADA el primer modelo TTS open source que genera texto y audio al mismo tiempo y sincronizados.

Lo loco: - 0… → tweet link

Developer Tools & AI Coding

@LinusEkenstam · 2026-03-11T18:33

Some fresh $400M at a $9B valuation. And Replit Agent 4.

Launching all this minutes before I start a 36 hour session of Flights from Asia to Europe is bad timing for me 😂 → tweet link

@ollama · 2026-03-10T21:29

Ollama can now run prompts on a schedule in Claude Code. Stay on top of work by setting automated tasks or reminders.

ollama launch claude

/loop Give me the latest AI news every morning

Examples in thread https://t.co/UtbaNFdbIv → tweet link

@karpathy · 2026-03-11T16:22

Expectation: the age of the IDE is over Reality: we're going to need a bigger IDE (imo).

It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It's still programming. → tweet link

@jezell · 2026-03-11T18:29

RT @OpenAIDevs: 📣 Technical lessons from building computer access for agents

Making long-running workflows practical required tightening t… → tweet link

@jezell · 2026-03-11T15:49

RT @initjean: introducing slopmeter

a cli tool to create a sharable nice looking graph to show off your Codex, Claude Code, or OpenCode us… → tweet link

@jezell · 2026-03-11T05:11

I told you video was coming to the Responses API soon: https://t.co/sLAqBWPbu1 → tweet link

@iamdevloper · 2026-03-11T08:16

RT @_benhowdle: 12 years of a free fake API, now it's a real backend you call with fetch().

Collections, auth, logs, AI generation. No bac… → tweet link

Local AI & Hardware

@levelsio · 2026-03-11T18:50

RT @perplexity_ai: Announcing Personal Computer.

Personal Computer is an always on, local merge with Perplexity Computer that works for yo… → tweet link

@FrameworkPuter · 2026-03-11T18:42

Don't let AI service providers dilute the meaning of "local". If it can't run offline on compute and data you actually own, it's not a Personal Computer. → tweet link

@ollama · 2026-03-11T03:30

RT @NVIDIARobotics: Need a quick guide to running OpenClaw locally on an NVIDIA Jetson? 🦞

Check out this walkthrough using @Ollama and a c… → tweet link

Infrastructure & Reliability

@karpathy · 2026-03-11T18:01

My autoresearch labs got wiped out in the oauth outage. Have to think through failovers. Intelligence brownouts will be interesting - the planet losing IQ points when frontier AI stutters. → tweet link

@jezell · 2026-03-11T16:12

Switched to blacksmith because github actions keep going down. blacksmith arm docker runners randomly started failing with OOM yesterday on the largest runners (same workloads work fine on local tiny macs), preventing 100% of our arm builds from working there.

Starting to feel like hosted CI providers just aren't worth it anymore. → tweet link

@jezell · 2026-03-11T16:29

RT @samlambert: AI has made it easier than ever to waste someone else's time → tweet link

Software Development & Engineering

@MatejKnopp · 2026-03-11T19:05

More fun with skia_dart. This time on Linux - graphite rendering into a wayland subsurface on background isolate while staying synchronized with Gtk window decorations. Seems like this finally might be getting somewhere. https://t.co/mPCxnNQfMj → tweet link

@iamdevloper · 2026-03-11T10:00

The moment you accept that "Works on my machine" can't be an acceptable deliverable, you've transitioned from a beginner to a real developer. But hey, at least we still have imposter syndrome to look forward to. → tweet link

@MatejKnopp · 2026-03-10T22:00

Private. Fucking. Named. Parameters. → tweet link

@jezell · 2026-03-11T08:02

RT @jkup: Source maps are a vital part of modern web development, but the lack of a standard made it impossible to add new features or impr… → tweet link

@iamdevloper · 2026-03-10T19:50

did you know there are people out there who receive emails and immediately just...respond to them

psychopaths → tweet link

Tech Industry & Recruiting

@FinansowyUmysl · 2026-03-11T09:38

Rekrutuję programistów w swoje pracy.

Zauważyłem ciekawą zmianę na przestrzeni ostatniego roku.

Coraz trudniej się wyróżnić.

Większość kandydatów ma dobre CV, dobrze wypada na zadaniach i rozmowach.

Zdarzało mi się rekrutować, na niższe stanowiska, osoby z ogromnym doświadczeniem, które kilka lat temu byłyby na znacznie wyższych pozycjach.

Wydaje mi się, że coraz więcej firm będzie odsiewać kandydatów w inny sposób niż rozmowy.

Zakładam, że wrócimy do czasów, kiedy będą się liczyć dodatkowe projekty, którymi możesz się pochwalić. → tweet link

@jezell · 2026-03-11T08:28

RT @tunguz: We've reached the point in software development where it's far far easier for a small cracked team to actually build a product… → tweet link

@TrungTPhan · 2026-03-10T23:50

RT @bearlyai: Former Dropbox CTO says rise of AI code has completely changed software engineer recruiting (vastly increases value of side p… → tweet link

@RealGeneKim · 2026-03-11T16:40

RT @dfinke: Jeffrey Snover says knowing too much stops people from trying AI.

This Friday I'll show you how to stop knowing and start doin… → tweet link

Open Source & Databases

@jezell · 2026-03-11T15:44

RT @lancedb: 1/6 Lance Blob V2 introduces multi-semantic blob storage under a single column type.

Instead of forcing all binary data into… → tweet link

@jezell · 2026-03-11T15:44

RT @matthewisabel: Increasingly common perspective on OSS collaboration based on operating mode of github/gh-aw https://t.co/3LM42U4qxd [→