Software Tech Radar

Garbage in, garbage out hits hardest with LLMs — poor data creates biased outputs, hallucinations, and legal risks.

Data prep consumes up to 80% of...

Four distinct approaches reveal clear trade-offs for persistent AI agent memory.

• Folder-based entities (people/, concepts/): Lowest complexity with...

Google Antigravity CLI delivers multi-agent coding directly in the terminal with multi-step reasoning, multi-file editing, tool calling, and...

DeepLearning.AI's new short course teaches developers how to extend agent workflows into multimodal generation with image and video agents that...

• Anthropic closing >$30B round at $900B+ valuation, surpassing OpenAI as top private AI startup.
• Revenue exploding from Claude Code tool, with Q2...

SuperClaude layers a modular command-based system over the Anthropic API, creating stateful agents that retain context across multi-step projects.

-...

Together AI just expanded its on-demand clusters and Dedicated Endpoints with a thousand H100s and H200s. This move highlights surging demand for readily available AI compute—worth tracking if you're budgeting or scaling GPU-dependent workloads.

Engineering teams now face a unified checklist across deployment, methodology, tools, security, and reliability.

• Platform: Agentbox unifies model...

Modelence spins up a full SaaS with MongoDB, validation, and automatic AI call tracking (latency, tokens, model, status) in 7 minutes—no logging code...

AI-augmented Playwright frameworks are proving transformative for non-technical QA teams by slashing maintenance costs while boosting coverage.

-...

LangGraph shifts RAG from fragile single-shot pipelines to autonomous loops that rewrite queries, evaluate results, reflect, and retry.

For customer...

Hybrid LLM Routing for Agents

• 🔥 OpenClaw Telegram Bot Demo: Demonstrates routing simple queries to local Ollama and complex queries to Claude...

Agents are moving beyond static tools toward autonomous self-improvement across two fronts.

• Hermes setups pair Obsidian with Google Drive to give...

A clear trend toward practical, step-by-step AI agent tutorials is emerging for developers.

• General agents follow a 9-step process: define purpose,...

Teams scaling AI agents from pilot to production consistently hit the same gaps.

• Start with organic experiments rather than top-down mandates;...

• Reinforcement learning unlocked 80-100+ tool-call agents and minimal harnesses scoring 76% on SWE-bench
• MCP secured consumer and enterprise wins...

Google I/O sessions reveal complementary paths for moving AI agents into production using shared tools.

• ADK and Antigravity deliver core primitives...

AI observability is the practice of making artificial intelligence systems transparent and measurable in AI deployment.

Enterprises are rapidly adopting zero-trust models as AI agents enter production, driven by risks like unintended actions and credential exposure.

-...

Route simple queries to local Ollama and complex ones to Claude via OpenClaw router in your Telegram bot setup to halve API spend. The approach uses a spec-driven prompt to intelligently classify and forward requests.