Software Tech Radar

Agentic Workflows & Orchestration

• 🔥 Asana Acquires StackAI: Asana acquired StackAI for $75M to extend AI Teammates with no-code agent execution...

Multi-Token Prediction achieves over 3x faster inference with zero quality loss by improving on speculative decoding, as shown in benchmarks on Gemma 4 31B (vLLM) and Qwen 3.6 27B (llama.cpp) reaching 132 vs 39 tokens/sec.

Retrieval has evolved past single-shot RAG into dynamic agentic pipelines and unified frameworks that handle heterogeneous sources.

• Agentic RAG...

Asana's $75M StackAI acquisition fills its key gap: AI agents that plan inside the platform but couldn't act across external ERP, CRM, and ITSM...

Production AI success hinges on infrastructure and operations, not model selection.

• Prototypes fail when model servers crash; the real burden is...

Two emerging platforms tackle long-running AI agents with different strengths.

• Google ADK targets enterprise workflows needing durable state...

This short video delivers a focused walkthrough of Chroma DB tailored for AI engineers.

• Core capabilities cover embeddings, multi-modal storage,...

The shift from agentic AI to physical systems is accelerating across expert forums, startup infrastructure, and accessible tooling.

• Panelists from...

Two distinct paths emerge in AI coding tools:

• Verdent AI deploys specialized agents for planning, frontend, backend, testing and optimization
• Zot...

Two recent tutorials reveal practical blueprints for moving multi-agent AI beyond prototypes.

• Identity layer first: Assign distinct identities,...

Oslo-based Cloudgeni raised €858k to scale AI agents that automatically detect, remediate, and prove cloud security and compliance. The platform targets the 99% of breaches from misconfigurations while eliminating manual DevSecOps overhead.

Multi-Agent Evaluation & Execution

• 🔥 AsyncTool Benchmark: AsyncTool evaluates asynchronous function calling capability under multi-task...

Will Mistral's Vibe give devs a reason to ditch Claude for agentic coding?

• Core pitch: AI agent purpose-built for work and code.
• Landscape shift:...

Four converging approaches show how structured knowledge and memory reduce hallucinations in AI agents.

• Knowledge bases must be AI-optimized with...

• Map full costs first: pilot spend of $200/month can reach $200k in production
• Choose model types wisely: instruction-tuned models cost 5-20x less...

Across emerging platforms, durable execution and governance layers are becoming non-negotiable for reliable AI agents.

• Itential stresses RBAC,...

AI agents face security threats primarily at the intersection of LLMs and traditional software tools, where old vulnerabilities like missing input...

Engineers now have streamlined paths across platforms for moving AI models and agents from quick tests to reliable production.

• Google Cloud Run...

Gemini 3.5 Pro is predicted to lead the next frontier model race, leveraging Flash's strong speed, grounding, and reasoning capabilities. This opens a...

AI engineers can leverage torch.compile to auto-generate Triton kernels and copy-paste them into production code for performance wins. The meta-pytorch/export-python experiment wraps this workflow behind a cleaner API.