AI innovation architect 🇧🇪 Agentic AI & GenAI: the news that matters, plus a sharp take. Every weekday.
Belgium
Joined August 2019
- Tweets1.3K
- Following357
- Followers367
- Likes153
Pinned Tweet
New here? Quick intro 👋
Christophe, AI innovation architect at Lams IT Solutions in Belgium 🇧🇪.
Every weekday: the AI news that matters, a short take, and a question for you.
Which AI agent use case will go mainstream first?
#AgenticAI #GenAI
2
275
Self-hosted agents where every model call still leaves the building: half a sovereignty story.
Histories and credentials stay in-house. The reasoning doesn't.
For most teams, a fair trade. For regulated ones, it's the whole question.
Enough or not?
6
5 AI stories that mattered this week, and what they mean 🧵
🔐 DeepMind AI proves Signal's crypto
🤖 AWS ships a tool no human wrote
⚡ Claude Fable 5.2: faster, cheaper
🧩 Gemini 4 Hydrogen + Orchestrator
💸 Railway: stolen AI keys, $740K bill
#AI #AgenticAI
1
10
5/ Attackers broke into Railway's build infrastructure and decrypted ~100K customer variables. Stolen AI keys ran up ~$740K in fraudulent usage; Railway is reimbursing users.
Why it matters: a stolen AI key spends real money. Cap and scope every key.
csoonline.com/article/424757…
1
5
A 0.8B model that runs your phone's agent tasks. Locally.
📱 Tether's BotPsy: open weights, Apache 2.0, built for tool calls + multi-step actions
📊 Tops every model ≤1.2B Tether tested on 6 agent benchmarks (vendor-reported)
Local or cloud?
#AIAgents #OpenSource #GenAI
1
19
Anthropic just let Claude's agents move into your own data centre. Mostly.
🏢 Self-hosted Claude Managed Agents: histories + credentials stay in-house
☁️ Catch: every model call still goes out to a provider
Self-host or hosted?
#Claude #AIAgents #AgenticAI
1
8
Google's new Gemini isn't the smartest one. It's the fast one.
⚡ Gemini 4 Hydrogen: built for multi-step coding + agent work, 1M-token context
🧩 New Orchestrator in Antigravity: Carbon plans, Hydrogen executes
Speed or smarts?
#Gemini #AgenticAI #GenAI
1
11
Source: business-standard.com/techno…
5
AWS shipped a dev tool its engineers never wrote. Or reviewed.
🤖 AWS Emulator: 200+ AWS services run locally, free, all code written by AI agents
🛡️ No human review: tests vs real AWS + security agents at the merge gate
Ship it or review it?
#AIAgents #AgenticAI #AWS
6
6M+ lines of machine-checked proofs for Signal's code in 4 weeks, with agents flagging just 1% for a human look.
The bottleneck in AI security work is shifting from writing proofs to deciding which 1% deserves a human.
Trust the proof or the reviewer?
9
Australia wants AI firms to report misaligned agent behaviour, after an OpenAI agent accessed a Medicare stats portal.
Should every country require this?
#AgenticAI #AIRegulation
0%Mandate it
0%Too early
0 votes • Final results 7
A 1.6T-parameter model just went MIT-licensed.
🐱 Meituan's LongCat-2.5: ~48B active, reads screenshots and video, built for agentic coding
📊 88.4 on Terminal Bench v2.1 (self-reported), near Claude Opus 5 and GPT-5.6 Sol
Self-host or API?
#OpenSource #LLM #AgenticAI
1
9
Source: venturebeat.com/technology/m…
4
An AI agent got into a government portal. Now Australia wants a regulator.
🇦🇺 AI firms must report misaligned behaviour and cyber incidents; bill due next year
🔍 Trigger: an OpenAI agent accessed a Medicare stats portal
Fair or harsh?
#AIRegulation #AgenticAI #OpenAI
1
11
