State of AI Engineering

What The State of AI Engineering Report Reveals About AI in Production

AI engineering has moved past experimentation, but are your observability practices keeping up? Datadog analyzed LLM telemetry from 1,000+ customers to reveal what's actually happening at scale: model adoption shifts, hidden token costs, and what the rise of agentic frameworks means for reliability. Download the report to benchmark your AI stack against production reality.

GitHub Actions Turned Into an Attack Tool ⚠️ - Attackers compromised GitHub repositories and abused GitHub Actions workflows to launch internet-wide scans for vulnerable cPanel and WHM servers. Instead of targeting developers directly, they weaponized GitHub's cloud infrastructure to perform automated reconnaissance. Read more.

Hugging Face Breached! - Hugging Face confirmed that one of its AI agents was compromised and gained unauthorized access to parts of its production environment. The incident was quickly contained, with no evidence of customer data exposure or model tampering. Read more.

AWS $1 Billion bill: An AWS billing incident caused estimated cloud costs to appear significantly higher than actual usage for many customers. The issue only affected cost estimates not real charges or running workloads and has since been resolved. Read more.

Why CRUD is dead?

For more than two decades, most Software-as-a-Service (SaaS) applications have followed the same basic model: CRUD Create, Read, Update, and Delete. Whether you're using Jira, Notion, HubSpot, or any other business tool, the workflow is largely the same. Users create records, update information, and remove data when it's no longer needed.

But AI is changing that model. Instead of simply storing information, modern software is beginning to take action on behalf of users, marking the start of a new era of semi-autonomous SaaS.

From Systems of Record to Systems of Action

Traditional SaaS applications act as systems of record. They store information but rely on humans to make every decision. AI-powered applications are shifting toward becoming systems of action, where software can analyze data, make recommendations, and even execute tasks automatically.

Imagine a finance platform that detects unusual spending, drafts messages to employees, and suggests budget changes without waiting for manual intervention. Or a project management tool that updates tasks, prioritizes work, and reorganizes backlogs by analyzing GitHub commits, Slack conversations, and pull requests. These systems don't just manage data they actively help complete work.

AI Copilots Are the First Step

We're already seeing this transition through AI copilots integrated into popular SaaS platforms.

Today's AI assistants summarize meeting notes, generate emails, detect fraud, and organize project discussions. While they still require human approval before taking action, they demonstrate how AI can reduce repetitive work without replacing users entirely.

Behind the scenes, these copilots rely on large language models combined with APIs that allow them to understand company data, reason about it, and perform predefined actions.

The Rise of Multi-Agent SaaS

The next evolution goes beyond a single AI assistant.

Future SaaS platforms will allow multiple AI agents to collaborate across different business applications. For example, a finance agent could identify a budget overrun, consult a project management agent to understand delays, retrieve meeting summaries from a documentation platform, and automatically generate a report for leadership.

Instead of isolated applications, software becomes an interconnected network of intelligent agents sharing context and coordinating work in real time.

Eventually, AI won't simply recommend actions it will execute them within predefined boundaries.

A CRM could automatically launch marketing campaigns, optimize messaging, and pause underperforming ads. Project management software could assign tasks, adjust deadlines, and resolve scheduling conflicts without manual updates. Finance platforms could forecast spending, enforce budgets, and negotiate vendor renewals using AI-driven workflows.

In this model, users become supervisors rather than operators.

The Future of SaaS

CRUD isn't disappearing it remains the foundation for storing and managing data. However, it is becoming the infrastructure beneath a much smarter layer of automation. The next generation of SaaS will focus less on recording information and more on completing work through AI agents that reason, collaborate, and act.

As businesses continue adopting AI, the companies that redesign their products around actions instead of records will be best positioned to define the future of enterprise software.

The Open-Source AI Shaking Up Silicon Valley - Moonshot AI's open-source Kimi model is emerging as one of China's strongest challengers to leading Western AI systems with impressive coding and reasoning capabilities. Its performance, lower cost, and open-weight release are fueling global interest and intensifying the AI race between the U.S. and China. Read more.

Reddit says bye bye HTML! - Reddit now requires authentication for Old Reddit, arguing its legacy HTML interface lacks the modern security stack needed to combat large-scale scraping. The article suggests the real motivation is making human-generated content harder for AI crawlers to harvest by forcing JavaScript-heavy rendering and authenticated access. Read more.

Buzz of the Week!

Semantic Caching

Semantic caching stores responses based on the meaning of a query instead of an exact text match. Rather than treating "How do I deploy a Kubernetes cluster?" and "What's the best way to deploy Kubernetes?" as different requests, it uses vector embeddings to recognize that both have nearly the same intent and serves a cached response. This significantly reduces LLM API calls, latency, and token costs while improving response times. Modern AI platforms, RAG systems, and enterprise chatbots use semantic caching to avoid repeatedly generating answers for semantically similar questions. The biggest challenge is choosing the right similarity threshold if it's too high, useful cache hits are missed; if it's too low, users may receive inaccurate or irrelevant responses.

Things that launched. Things that went viral. Things you'll pretend to try.

dekit

dekit lets you run and manage multiple development processes in a single terminal workspace.

pluto

pluto detects deprecated Kubernetes APIs before upgrades break your clusters.

buzz

buzz transcribes and translates audio offline on your personal computer. Powered by OpenAI's Whisper.

Build Braincells, Not Just Features

This weekend’s read: Which AI ?

This week’s watch: What if Earth is Type 3 civilization.

Meanwhile…