This Substack post outlines an AI roadmap for 2026, urging leaders to move beyond hype toward actionable plans. It argues that the wave of "agentic AI" will hinge on integrating planning, observability and human‑centred guardrails into systems design. Key themes include rapid iteration, rigorous testing of agent workflows, and focusing on user needs rather than model benchmarks. (The post wasn’t fully accessible in our environment, so this summary is based on public overviews and trends.)
Rocklin argues that experienced developers should embrace AI because it makes coding more enjoyable and lets them focus on higher‑order thinking. He lists “big ideas” such as minimising interruptions and climbing the abstraction hierarchy, and offers practical tips like using structured hooks when giving instructions to AI agents. At the same time, he acknowledges valid concerns: large models can generate junk code, blind reliance can erode understanding, reviewing AI output may be slower than writing from scratch, and AI workflows can feel dehumanising. The piece ends by urging balanced adoption and continuous learning.
Huryn’s New‑Year newsletter notes that the mainstream narrative about agentic AI lagged reality in 2025 and that by late 2025 reliable scoped workflows emerged. He argues the focus in 2026 will shift from building bigger models to designing entire AI systems — including observability, governance and protocols. Concepts like “vibe engineering” (prioritising design and testing over just coding prompts) and spec‑driven development will be key.
This session covers Lyft’s feature‑store architecture and its evolution over five years. The platform centralises feature engineering, provides low‑latency access and high‑throughput processing, and enables consistent features across models. Speakers explain how the store optimises feature management at scale and share lessons on performance, developer experience and long‑term evolution.
Uber describes its journey from a monolithic monitoring system to a cloud‑native observability platform. Built on an open‑source stack (Telegraf, Prometheus, Thanos, Grafana, Kibana) and deployed on Kubernetes, the new platform monitors Uber’s corporate network across regions. Key features include modular microservices, dynamic configuration (via a “Dynamic Config App”) that adapts to network changes, and global deployment for low‑latency measurements. The system emphasises resilience (automatic restarts) and scalable, high‑quality data.
Nóbrega introduces retrieval‑augmented forecasting (RAF), where a model searches a database of historical time‑series segments to find similar patterns and uses them to improve predictions. Instead of relying solely on model parameters, RAF adds an explicit memory access step — for zero‑shot scenarios, rare events or evolving seasonal trends. The article explains how queries are embedded, how similarity search retrieves relevant past episodes, and why this helps models handle anomalies.
Ty is a Rust‑based Python type checker and language server that’s 10×‑60× faster than mypy and Pyright. Designed for incrementality, it recomputes diagnostics in milliseconds, supports advanced type features (intersection types, reachability analysis) and integrates with VS Code and other editors. Astral plans to stabilise ty in 2026 and extend it to power semantic tools across its ecosystem.
A desktop app for Anthropic’s Claude Code environment. It lets developers run multiple Claude sessions locally, each in its own git worktree, and copy files from .gitignore automatically. The app also launches secure cloud sessions and supports custom environment variables. It’s currently in preview and integrates seamlessly with Claude Code on the web.
Sutskever discusses why the era of endlessly scaling models is giving way to research into new architectures, interpretability and safety. He argues that bigger models alone won’t unlock the next leaps in capability and that research‑driven innovation will define AI progress.
A concise talk on turning AI prototypes into production value. Speakers outline how to move from experimentation to scalable deployment, emphasising robust data pipelines, governance and alignment with business objectives.
Confluent’s annual predictions webinar examines new technical realities of AI in 2026 and offers guidance on future‑proofing data ecosystems. Speakers will highlight why traditional databases can’t handle the query surge and present architectures optimised for speed, scale and resilience. The talk is aimed at CTOs, data architects and platform engineers and is based on insights from Confluent’s 2026 Predictions Report
Xebia webinar on how Model Context Protocol (MCP), spec‑driven development and conversational data interfaces enable AI‑assisted data‑lakehouse engineering