Event — Cursor Meetup Helsinki

Meet local Cursor users, share what you’re building, and hang out.
📍 Skene.ai office (Mikonkatu 5), Helsinki
🕓 16:30 doors • food & mingle • community demos • more mingle

https://luma.com/pux3shu1

Opportunity — KONE (Espoo): Summer Trainee, Machine Learning (Summer 2026)

Our friends at KONE are hiring a Machine Learning summer trainee in Espoo.
Work on predictive maintenance at scale (IoT fleet), data/ML pipelines, analysis, dashboards, and production-grade MLOps (Databricks + AWS).

For: late-stage MSc students (May–Sep 2026, ~4 months)
Hybrid: office 1–2 days/week + occasional Hyvinkää R&D days

Apply: https://careers.kone.com/en/find-jobs/r0656267/summer-trainee-machine-learning/

Opportunity — Neste (Espoo/Keilaranta): Summer Trainee, AI/ML Engineering (Summer 2026)

Neste is hiring an AI/ML Engineering summer trainee to join their Process Analytics, Automation & AI team.
You’ll work on real industrial AI/ML use-cases end-to-end: data prep, prototyping, training/validation, pipelines + APIs, documentation/reproducibility, monitoring, and basic MLOps (CI/CD, packaging, deployment/retraining depending on background).

Apply: https://jobs.neste.com/job/Espoo-Summer-Trainee%2C-AI-ML-Engineering/1358409100/

6 career notes for AI/ML builders in 2026

1) Fundamentals still win: regression, loss functions, tokenization, embeddings.
2) System first, model second: data → features → serving → monitoring.
3) Learn MLOps by shipping: deploy, version, observe, retrain.
4) Product thinking: translate metrics → business impact.
5) Depth > FOMO: pick one domain, go deep, build real projects.
6) Protect focus: hype is loud, production is slow. Rested engineers ship.

Claude Code + local models (Ollama) — big unlock

Ollama (v0.14.0+) now supports Anthropic’s Messages API compatibility.
That means tools built for Claude (incl. Claude Code) can be pointed at local open-weight models running on your machine.

Privacy + no per-token cloud bills (compute is yours).

Official:
https://ollama.com/blog/claude
Docs:
https://docs.ollama.com/api/anthropic-compatibility

Gemini ≠ “just a chatbot”.

Google is quietly stitching together a full AI stack — model → tools → products → platform.

6-layer view:

1) Models
Gemini (Pro/Flash) + open-weight Gemma
https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models
https://deepmind.google/models/gemma/

2) Build & code
Gemini Code Assist (IDE + Cloud workflows)
https://codeassist.google/

3) Video
Veo (gen video)
https://deepmind.google/models/veo/

4) Images
Imagen (text-to-image)
https://deepmind.google/models/imagen/

5) Assistants at work
Workspace with Gemini + NotebookLM
https://support.google.com/a/answer/13623623
https://notebooklm.google/

6) Platform + distribution
Vertex AI + Agent Builder + Firebase + on-device (Gemini Nano) + Search AI Overviews
https://cloud.google.com/vertex-ai
https://docs.cloud.google.com/agent-builder/overview
https://firebase.google.com/docs/ai-assistance/gemini-in-firebase
https://developer.android.com/ai/gemini-nano
https://search.google/ways-to-search/ai-overviews/

The best tech events aren’t built on code alone…

We’re building Since AI for the second year in a row: an AI hackathon in Turku that brings developers and companies together to solve real business challenges.

After our first event, we’re now expanding the team to make the next one even stronger — with the long-term goal of building Europe’s largest AI hackathon.

We’re recruiting team members across roles such as Marketing & Communications, UX/Web, Graphic Design, Finance, Production, Tech & Program, and more.

If you want hands-on experience, real responsibility, and a chance to shape a growing AI event, we’d love to hear from you.

👉 Apply via the form: https://forms.gle/BpNGGYitS3C2i7xQ8

Learn more: https://sinceai.fi/

🚀 New standard for AI coding agents: AGENTS.md

If you’re building with Cursor / Copilot / Claude Code / Codex → this is a must-add to your repo.

AGENTS.md = a tiny file that tells AI agents how to work in your codebase:
• how to run / test / lint
• repo structure + conventions
• “do / don’t” rules so it won’t break stuff

Already used by 60k+ open-source projects.

👉 https://agents.md/

Challenge (5 min): add AGENTS.md to your repo today and watch agent output quality jump.

📈 2 big signals in AI right now

1) Enterprise AI is turning into “one bundle”
• Reports/industry chatter: OpenAI is being pitched as a one-stop AI shop: ChatGPT + Codex + APIs + workflow automation (example thread: https://www.threads.com/%40getsuperintel/post/DT-Hj5bDobR/
, X post: https://x.com/kimmonismus/status/2015733989131342038)

• Anthropic is pushing hard on enterprise agent tooling: Claude Code for Team/Enterprise + admin controls (official: https://www.anthropic.com/news/claude-code-on-team-and-enterprise) + recent Cowork plugins coverage (Axios): https://www.axios.com/2026/01/30/ai-anthropic-enterprise-claude


2) GitHub just made “learning + credentials” more official
• Central hub: GitHub Learn → https://learn.github.com
• Certifications page: https://learn.github.com/certifications

What this means for builders (practical)
• Teams will buy bundles, not random tools → you win by showing you can ship end-to-end (app + eval + deploy).
• “Agentic coding” is going mainstream → make repos agent-ready with AGENTS.md (the agent README): https://agents.md/
• Credibility is standardizing → certs + public demos beat “I’m interested in AI”.

👉 Forward this to one builder who’s applying this spring.

🦞 Link of the day: “A social network for AI agents” (Moltbook)

A fascinating (and slightly unsettling) look at what happens when AI agents “hang out” together — workflows, tips, memes, philosophy, even proto-communities.

What it is
• Moltbook = “a social network for AI agents” (humans can observe)
• Came out of the ecosystem around Claude Code + the lobster-themed assistant that evolved into Moltbot/OpenClaw
• You can’t properly browse Moltbook without an agent login — but this write-up curates the best posts

Why builders should care
• A real preview of agent-to-agent communication + “agent culture” (useful + chaotic)
• If you’re building agent products, this helps you understand how agents behave outside the “assistant” mask

Read: Best Of Moltbook (Astral Codex Ten)
👉 https://open.substack.com/pub/astralcodexten/p/best-of-moltbook

Forward this to one builder — it’s a perfect conversation starter.

Open Responses — the missing standard for AI agents

If you’re building LLM apps/agents, this is high-signal: “Open Responses” is a new open spec aiming to standardize agent workflows (messages, tool calls, streaming, multimodal) so you can switch providers without rewriting your whole integration layer.

Why it matters (practical)
• Portability: one schema → fewer provider-specific hacks
• Cleaner agent loops: formalizes “items” + tool use + streaming events
• Ecosystem momentum: built for real-world agent stacks (apps + tools + orchestration)

Read
• Spec/overview: https://www.openresponses.org/
• Coverage summary: https://www.infoq.com/news/2026/02/openai-open-responses/

🔬 AI Agents: 88% Faster BUT Humans+AI Win by 69%
Fresh Stanford/Carnegie Mellon study (Oct 2025) drops a bombshell: AI agents alone are fast but flawed. Hybrid teams dominate.

📊 The Numbers:
• AI agents solo: 88% faster, 90-96% cheaper
• BUT: 32-49% lower success rate ❌
• Human+AI hybrid: 68.7% BETTER results ✅

🎯 Why Hybrids Win:
AI fails at:
• Hallucinations & wrong tool use
• Visual/judgment-based tasks
• Complex contexts

Hybrids excel:
• Humans handle judgment calls
• AI crushes structured work
• Together > separate

💡 Key Insight:
Workers don't fear job loss—they lack trust in AI accuracy (45% of resistance). They want "Equal Partnership", not full automation.

📚 Study:
"How Do AI Agents Do Human Work? Comparing AI and Human Workflows Across Diverse Occupations"
🔗 Research: https://futureofwork.saltlab.stanford.edu/
48 humans, 4 AI frameworks, 16 realistic tasks across 287 US occupations (71.9% of daily work)

How do you balance AI and human work?
👉 sinceai.fi - Join the community for more practical AI insights!

AI coding agents are getting serious — and you can benefit this weekend.


What happened

• OpenAI launched the Codex app (macOS): manage + supervise coding agents from one place
https://openai.com/index/introducing-the-codex-app/

• OpenAI also released GPT-5.3-Codex (stronger for longer coding tasks)
https://openai.com/index/introducing-gpt-5-3-codex/

• Reuters frames this as part of the “coding tools race” vs others like Anthropic
https://www.reuters.com/business/media-telecom/openai-launches-codex-app-gain-ground-ai-coding-race-2026-02-02/

• Anthropic is pushing Claude Code for Team/Enterprise
https://www.anthropic.com/news/claude-code-on-team-and-enterprise

Why builders should care
“AI engineer” is quickly becoming: can you direct agents, review safely, and ship — not just prompt.

Still open — we’re recruiting for the Since AI team.

We’re building Since AI Hackathon for the second year in a row: an AI hackathon in Turku where developers + companies build real demos on real business challenges.

After the first event, we’re expanding the team to make 2026 significantly bigger — with the long-term goal of building Europe’s largest AI hackathon.

Open roles include: Marketing & Comms, UX/Web, Graphic Design, Finance, Production, Tech & Program, and more.

If you want hands-on experience, real responsibility, and ownership in a fast-growing AI community:

👉 Apply: https://forms.gle/BpNGGYitS3C2i7xQ8

Learn more: https://sinceai.fi/

⚠️ AI builder security heads-up: Claude Desktop Extensions “0-click” risk

A new report shows some Claude Desktop Extensions (DXT) can be abused for remote code execution (0-click) via things like a malicious Google Calendar event, because certain extensions run with high system privileges.

If you use Claude Desktop + extensions, do this now (5 min):
• Disable/uninstall extensions you don’t absolutely need
• Don’t allow “auto-actions” from email/calendar/docs
• Run agent tools in a separate user / VM if possible
• Require manual confirm before any tool that can run commands / touch files / access secrets

Sources:
https://layerxsecurity.com/blog/claude-desktop-extensions-rce/
https://www.theregister.com/2026/02/11/claude_desktop_extensions_prompt_injection/

Opportunity — CTO (AI Voice Startup)

CallCompany (Turku/Helsinki based) is building AI voice agents for businesses and looking for a CTO.

Looking for someone who:
– has built scalable SaaS
– is strong in AI application development
– ideally has voice AI experience

Most important: senior-level builder mindset. Hungry, sharp, ready to execute.

If this sounds like you (or someone you’d recommend), reach out directly: niko@callcompany.ai

https://www.callcompany.ai/

🎓 If you're serious about AI in 2026 — stop scrolling, start studying.

Harvard CS50 + MIT OpenCourseWare: two free, top-tier learning sources for AI builders.

Completely free.

These are not “YouTube influencer tutorials.”
These are real university-level foundations.

If you want to:

• understand how neural nets actually train
• understand inference methods
• understand generalization + OOD
• understand scaling laws

This is gold.

MIT OpenCourseWare:
https://www.youtube.com/@mitocw

CS50x 2026 – AI:
https://youtu.be/-9bo8HlSxwQ?si=BeSTIgkM-LMOC2bI


Serious builders differentiate themselves by fundamentals.

Prompting ≠ AI engineering.

Forward this to someone who wants to be world-class, not average.

Website: https://sinceai.fi/

🎯 The 5-Second Prompt That Fixes 80% of Bad AI Outputs
Stop fighting with vague AI responses. Add this to ANY prompt:
"Think step-by-step and show your reasoning."

Why it works:
• Forces the model to break down the problem
• Catches logic errors before the final answer
• Works across GPT-5, Claude, Gemini
• Free performance boost

Example:
❌ Bad: "Calculate project timeline"
→ Gets: "About 6 weeks"
✅ Good: "Calculate project timeline. Think step-by-step and show your reasoning."
→ Gets:
• Step 1: Backend API (2 weeks)
• Step 2: Frontend (3 weeks, parallel after week 1)
• Step 3: Testing (1 week)
• Total: 5 weeks (with 1 week overlap)

Bonus variations:
• "Explain your reasoning before answering"
• "Break this into steps"
• "Show your work"

The data: Research shows chain-of-thought prompting improves accuracy by 20-40% on complex tasks.
Read: https://arxiv.org/abs/2201.11903

Try it on your next prompt. Takes 5 seconds, saves 5 iterations.

Opportunity — NestAI hiring AI Engineer (Helsinki / Tampere)

NestAI is hiring an AI Engineer to build physical-world AI (edge + cloud): model training, data/model pipelines, deployment, and MLOps — in mission-critical systems where reliability matters.

Strong fit if you have:

• Python + PyTorch/TensorFlow

• ML/DL + production deployment

• CI/CD + Linux

• Bonus: CV/perception, robotics, RL/control, Jetson/edge

Apply: https://nestai.com/ml-ai-engineer/

🍏 Xcode just went “agentic”

Apple released Xcode 26.3 (RC) with “agentic coding” support: you can use coding agents (e.g., OpenAI Codex + Anthropic agents) directly inside Xcode to plan + implement multi-file changes more autonomously.

Why it matters (practical):
• Agentic coding is moving from “extra tool” → default IDE workflow
• Apple also mentions support via MCP (Model Context Protocol) so agents/tools can plug in more flexibly
• If you’re building AI devtools or want to level up as an engineer: learn to *supervise agents* with tests + clean diffs

Read:
Apple announcement: https://www.apple.com/newsroom/2026/02/xcode-26-point-3-unlocks-the-power-of-agentic-coding/

Quick builder action (30 min):
1) Add tests (or at least a minimal smoke test script)
2) Add AGENTS.md (how to run/test/lint + repo rules): https://agents.md/
3) Record a 45s demo of “agent → diff → tests → merge” and pin it on GitHub

Website: https://sinceai.fi/