Gemini can now prep a rideshare or grocery order, though you’ll have to submit the order yourself. | Image: Google
Google's Gemini AI is getting one step closer to being more like an actual assistant. Starting with some Pixel 10 phones and the Samsung Galaxy S26 series, Gemini will be able to hail an Uber or put together a DoorDash order on its own.
It's called task automation, and it starts with a prompt to Gemini - something like "Get me an Uber to the Palace of Fine Arts." Gemini then launches the app in a virtual window on your device and goes through the process step-by-step. You can watch it all happen, with options to stop the automation or take control if necessary, or just let it run in the background while Gemini does its thing. The assistant …
TL;DR: Today, we’re releasing a new episode of our podcast AI & I, where Dan Shipper sits down with Alex Mathew, a 17-year-old student from Alpha High School, an unconventional AI-forward high school in Austin, Texas. Watch on X or YouTube, or listen on Spotify or Apple Podcasts.
Was this newsletter forwarded to you? Sign up to get it in your inbox.
Depending on whom you ask, AI is either the best or worst thing that can happen to the next generation. The arguments come from educators, venture capitalists, op-ed writers, and anxious parents—but rarely from the young people in question.
On this episode of AI & I, Dan Shipper sat down with one: Alex Mathew, a 17-year-old high-school senior at Alpha High School in Austin, Texas.
Alpha School, a rapidly expanding network of kindergarten through grade 12 private schools, is not without controversy. Inside Alpha High School, there are no traditional teachers, all academic content is delivered through an AI-powered platform, and the adults in the classroom, known as “guides,” focus solely on supporting the students emotionally and keeping them motivated to learn. The students have two- to three-hour learning blocks every morning and spend the rest of the day going deep on a project in an area they care about, spanning art, sport, life skills, and entrepreneurship.
JMESPath query support for filtering and transforming JSON output
Deploy to Azure App Service deployment slots directly from azd
Automatically install azd extensions in dev containers
New --subscription and --location flags for azd provision and azd up
Extension version requirements with requiredAzdVersion
Remote build support for Azure Functions Flex Consumption plans
We want to hear from you!
Help us shape the future of azd by sharing your experience. We’re conducting user research to better understand how you’re using the Azure Developer CLI. Sign up to participate and make your voice heard!
New features
Query and output
azd now supports JMESPath queries, giving you the ability to filter and transform JSON output directly in the terminal.
JMESPath query support: The new --query flag lets you filter and transform JSON output from any azd command using JMESPath expressions—great for scripting and automation. Query support also covers Message-type outputs, so filtering works consistently across all commands. For a deep dive, see our 1-min blog post: JMESPath query support in azd. Thanks @scottaddie for the contribution! [#6664][#6735]
Deployment and infrastructure
Several improvements streamline the deployment experience, from deployment slot support to more flexible provisioning workflows.
App Service deployment slot support: azd now deploys directly to Azure App Service deployment slots without extra scripts. For details, see our 1-min blog post: Deploy to App Service slots with azd. [#6627]
Subscription and location flags: New --subscription and --location flags for azd provision and azd up allow overriding defaults per command without changing global or environment configuration. [#6777]
Remote build for Functions Flex Consumption: New remoteBuild configuration option enables remote builds when deploying to Azure Functions Flex Consumption plans, avoiding local build requirements. [#6748]
Extensions
Required azd version for extensions: Extensions can now specify a requiredAzdVersion field, ensuring users run a compatible version of azd before the extension loads. [#6747]
Dev containers
Automatically install extensions in dev containers: The azd dev container feature now supports an extensions option, so you can specify azd extensions to install automatically when a dev container is built. For details, see our 1-min blog post: Automatically install azd extensions in dev containers. Thanks @stuartleeks for the contribution! [#6460]
AI and automation
AI coding agent auto-detection: azd now detects when it’s running inside an AI coding agent and skips interactive prompts automatically, so automated workflows don’t hang. [#6633]
Bugs fixed
CLI and terminal
Fixed arrow keys displaying as escape sequences in the Ghostty terminal emulator. [#6739]
Fixed missing configuration keys not appearing in azd config options. [#6619]
Deployment and infrastructure
Fixed remote build returning 404 errors when Azure Container Registry is in a different resource group than the target resource. [#6766]
Fixed subscription cache being overwritten when switching between subscriptions. [#6770]
Fixed azd init to fail fast when required values are missing in non-interactive mode. [#6779]
Extensions and configuration
Fixed duplicate azd-service-name tag error during provisioning. [#6674]
Improved provisioning error messages with more actionable guidance. [#6690]
Improved error messages when provisioning fails due to region SKU unavailability, including suggestions for alternative regions. [#6771]
Improved error classification for better telemetry and troubleshooting. [#6803]
Improved delegated authentication messaging during interactive login flows. Thanks @scottaddie for the contribution! [#6808]
Added soft-delete conflict detection hints when resource creation fails due to previously deleted resources. [#6810]
Refactored container helper for improved maintainability. [#6649]
New docs
The Azure Developer CLI documentation continues to expand:
Layered provisioning (February 12): New guide covering layered provisioning patterns for managing shared and environment-specific infrastructure in azd projects. Learn more
Full-stack deployment guide (January 28): New content on deploying full-stack applications with azd, covering frontend, backend, and database components. Learn more
Multi-tenant authentication guidance (January 29): Updated guidance on configuring azd for multi-tenant authentication scenarios. Learn more
New templates
The Awesome azd template gallery picked up 10 new community templates this month. Thank you to every contributor who shared a template—you make it easier for the whole community to get started on Azure.
Deploy an MCP server with OAuth 2.1 authentication and On-Behalf-Of (OBO) flow to Azure Container Apps with secretless deployment using Federated Identity Credentials.
One-click azd template that deploys a Blazor Server chat app on Azure App Service integrated with Azure Foundry, VNet, managed identity, and Application Insights.
Get started with Kubernetes Event-driven Autoscaling (KEDA) scale rules in AKS. Includes optional support for Azure Container Registry, Managed Grafana, and App Routing.
Route requests to multiple AI services (Microsoft Foundry and Google Gemini) through a single endpoint with authentication, load balancing, and token limiting.
Deploy governance policies for Microsoft Foundry, M365 Copilot, and Fabric. Covers Purview Data Security Posture Management (DSPM), data loss prevention (DLP), sensitivity labels, and Defender for AI.
New to azd?
If you’re new to the Azure Developer CLI, azd is an open-source command-line tool that accelerates the time it takes to get your application from local development environment to Azure. azd provides best practice, developer-friendly commands that map to key stages in your workflow, whether you’re working in the terminal, your editor or CI/CD.
When OpenAI unveiled GPT-5.3-Codex in early February 2026, it marked a significant evolution in the Codex product line — not just as a coding assistant, but as an autonomous agentic work partner. Where earlier generations of Codex focused on generating high-quality code snippets and assisting developers with long-running tasks, version 5.3 is positioned as a more capable collaborator that blends coding prowess with professional reasoning and interactive execution. OpenAI is pushing Codex toward acting more like a fellow developer — one that can take direction mid-task, maintain context over extended workflows, and tackle complex real-world development jobs end-to-end.
TL, DR: I built the same Uno Platform app with Codex 5.3 & 5.2 - Follow along the journey ..
Why do .NET developers care? Well, the developer chops Codex 5.3 brings to the table apply very much to the realities of enterprise app development scenarios. Add to that the nuances of running apps across platforms from single shared codebase, the benefits of a matured Agentic partner which can reason & churn out solid code, become obvious.
Uno Platform is the most flexible open-source platform for modern cross-platform .NET development, complete with enterprise-grade AI and visual design tools. Paired with Uno Platform Studio, .NET developers can elevate productivity with runtime visual designers and dependable AI Agents/MCP tools for contextual AI intelligence – all towards building apps from any OS/IDE and any AI Agent, to run on mobile, web, desktop or embedded devices.
For .NET developers building cross-platform apps, AI tooling in Uno Platform works with any Agentic workflows, including GPT-Codex. Let’s take a closer look as to how Codex 5.3 fares against 5.2 for Uno Platform .NET apps – we’re in hands off mode pitching AI Models against each other.
So yeah, Codex 5.3 is better and faster than Codex 5.2 in almost every way. But does 5.2 still have some deeper reasoning tricks up its sleeves? Let’s check out the developers experience.
Uno Platform Experience with MCPs
Uno Platform MCP Servers/Tools work really well with Codex – AI Agents can not only generate code, but use the tools to also verify app fuunctionality. To put the two versions of Codex to a fair test, how about we ask them to build the same app with the same exact prompts/tools?
Overall Prompt:
Build a me a car dashboard UI
Show a map with route overlay
Show navigational overlay on the map
Show dynamic lane visualizing our car
Show live traffic going by/overtaking in lanes
Show overlay to change Seat settings
Show overlay to control A/C with airflow
Show Media info that is playing now
Show overlay to control what Media is playing
Changeable UI data bound to overlay info
Additional things handed to Agents:
/ Couple of car dashboard screenshots for inspiration – likes of Tesla, BMW & Kia.
/ Images for top of car for lanes & seat graphic
/ Uno Platform Docs MCP for best practices
/ Uno Platform App MCP with app interativity tools for verification
The goal was to see what each Codex can do on their own – Fire off prompts in CLI and go hands-off.
Here’s what GPT-5.2-Codex built:
Here’s what GPT-5.3-Codex built:
Let’s look at the comparison – this is mostly anecdotal from a developer’s perspective:
Criteria
Codex 5.2
Codex 5.3
Overall outcome
Performed well on the overall task
Performed quicker on the overall task
Final result
Fully functional Uno Platform cross-platform app
Fully functional Uno Platform cross-platform app
Time to complete
~35 minutes
~20 minutes
Mapping approach
Used a real map
Did not use a real map; rather plotted names on canvas
Map awareness
Did not initially understand which part of the map represented what
Not applicable
Route overlay accuracy
Initially drew route over large bodies of water
Did not have this issue
Fixes required for route
Needed extra prompts to correct route overlay over water
Not applicable
Traffic labeling
Initially labeled traffic incorrectly
Canvas overlay did not need accuracy
Moving vehicles in center lane
Required extra prompt to prevent vehicles overlapping our car
Required extra prompt to prevent vehicles overlapping our car
Mapping overlay data binding
Needed fixes to properly bind overlay data changes to UI
Quicker and more accurate in binding overlay data changes to UI
Climate control airflow visualization
Took a literal and bold approach to represent airflow around dashboard
Took a safer approach using moving dots with variable velocity to represent airflow
Overall, both Codex 5.2 & Codex 5.3 performed admirably – entire functional UI built with AI and tested with Uno Platform MCP tools. While Codex 5.3 was substantially quicker, the deeper reasoning and brave more realistic approach taken by Codex 5.2 is to be appreciated.
Comparing Codex Agents
In many ways, developers today are spoiled for choice. Both Codex 5.2 and Codex 5.3 represent the cutting edge of AI-assisted software development, and it’s hard to go wrong with either. Each model brings a distinct personality to the table, and understanding those nuances helps teams pick the right tool for the moment rather than declaring a single universal winner.
Codex 5.2 often felt bold and exploratory — a model willing to take calculated risks, reason deeply through ambiguous problems, and push toward creative or non-obvious solutions. For developers tackling complex architecture, experimental ideas, or problems that benefit from heavier reasoning, 5.2 proved to be a remarkably capable partner.
Codex 5.3, by contrast, refines the experience. It is faster, more consistent, and noticeably more robust as an end-to-end coding agent. The gains in execution-style benchmarks, workflow fluency, and responsiveness make it particularly well suited for real-world development loops — writing code, iterating, fixing, and finishing tasks with fewer stalls and less friction.
Ultimately, this isn’t a story of replacement but of progression. Codex 5.2 showcased strong reasoning and risk-tolerant problem solving, while Codex 5.3 builds on that foundation with speed, reliability, and agentic strength. Developers truly have a plethora of riches — and whichever path they choose, they’re backed by some of the most capable coding models ever built.
Ready to Build with AI-Assisted Workflows?
Sign up to Uno Platform Studio for free and set up Uno MCP and App MCP in minutes. Start building cross-platform apps with AI agents that actually understand your code.
TX Text Control now offers a private NuGet feed with enhanced support for automated build pipelines. For organizations already using Azure Artifacts, that option remains fully available. This guide covers both approaches, helping you choose the best fit for your team.