THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.
All briefings

Follow the change ·

The day,
in perspective.

What happened in AI and robotics — and the change behind it. A reading path through 42 stories from BIG CHANGE.

Ordered using RADAR attention, story opens and likes. Excerpts from our published reporting; open a story for the full context and sources. Dates in UTC.

guide

Verify one OpenAI Agents SDK tool boundary before release

Test one existing Python Agents SDK FunctionTool against approved and denied arguments, then review an authorized staging effect, trace and usage for a bounded accept-or-hold decision.

The big change

What changed: The SDK's function-tool input guardrail can reject a proposed call before the selected Python tool body executes. This recipe binds that gate to the existing agent's code, configuration and dependency state.

Why it matters: A scripted allowed call and same-schema denied call can establish whether the local SDK path executes the selected tool. Only an independently observed staging effect can establish what the real action did.

What to watch: The owner must see the actual trace, effect and output, plus usage evidence for every provider response. An SDK token total of zero can also mean the provider omitted usage; missing evidence ends in HOLD.

Read the full story
—
Back to the briefing
guide

Build an agent feedback loop and check if it improves

Use Promptfoo, an existing task agent and approved fixtures to compare a baseline with one loaded lesson on development and held-out cases.

The big change

What changed: The SDK and Promptfoo can be joined around one existing agent, while an application-owned registry loads an approved lesson into its exact instructions.

Why it matters: Four bound development and held-out result sets let the task owner see individual fixes, regressions and evaluation shortcuts before retaining the change.

What to watch: The integration and real-task oracle require local implementation and testing. Results, traces and billing can contain sensitive data, and no improvement or cost has been measured here.

Read the full story
—
Back to the briefing
guide

Set up a managed database for an AI-assisted web app

Choose a managed Postgres path for an existing app, move a reviewed migration to the right Supabase development project, test per-user access and plan production recovery.

The big change

What changed: A database setup task becomes a traceable path from app requirements and versioned SQL to a named development project, migration history and schema readback.

Why it matters: A local migration or working screen cannot show that the managed app points at the right database or that cross-user operations are denied.

What to watch: Use two real permitted test identities and inspect responses and row state. Record the production authorization and recovery evidence separately.

Read the full story
—
Back to the briefing
guide

Release an owned ADK agent through a checked Agent Runtime revision

Move an existing ADK agent to a reviewed Agent Runtime revision with authenticated MCP access, verified users and sessions, approved traces and a traffic recovery route.

The big change

What changed: Google’s Agent Runtime can host an ADK agent as a managed reasoningEngine, with Agent Identity credentials for its outbound tool and active revisions for controlled traffic.

Why it matters: The tool’s API key identifies the agent to the MCP server. It does not establish which person submitted a request. An authenticated ingress and protected session-owner record have to make that decision before the agent runs.

What to watch: Offline evaluation needs approved content-bearing traces. “Always latest” traffic can expose an updated revision before review, while an archived revision cannot be restored.

Read the full story
—
Back to the briefing
guide

Build a booking and scheduling app with an AI coding agent

Give an AI coding agent a complete one-resource booking app brief, then check identity, contested slots, persistent history and requested email before real appointments.

The big change

What changed: An AI coding agent can take a complete application brief and produce the interface and server code. The owner can specify the booking rules, database constraints and tests before it writes code, using Replit Agent's Plan mode to review the work.

Why it matters: For a small service business, the useful outcome is a booking record that survives reloads and competing requests. A visible open slot is only an invitation to try; the database transaction must decide who gets it. The owner still controls enrollment, real availability and customer contact.

What to watch: A working sign-in screen or confirmation email does not establish a safe booking system. Check the generated app's contested-slot behavior, time-zone interpretation, public authentication paths and signed email callback on its actual host before accepting real appointments.

Read the full story
—
Back to the briefing
guide

Run Hemmingway-1 locally and check a first response

Load a pinned Hemmingway-1 checkpoint on a suitable Linux GPU, then check a first local response through vLLM or a separate Transformers script.

The big change

What changed: The publisher makes a writing-focused checkpoint available for local inference. A reader can send a prompt to their own server or script while keeping the model revision fixed.

Why it matters: This gives authorized users control over where the generation request runs. It also puts GPU setup, storage, license checks and response verification in their hands. The local request in this guide does not call a hosted model API; installing packages and downloading weights still use the network.

What to watch: The useful result is a nonempty new assistant response from the reader's machine. That proves this setup generated text, while the suitability of a message still depends on reading and checking it. Runtime support for the model architecture is documented, but this exact checkpoint has not been run by BIG CHANGE.

Read the full story
—
Back to the briefing
guide

Add sign-in and protected data to an existing React app

Use a coding agent to add Supabase email authentication to an existing React/Vite app, then check real callbacks, enrollment and per-user data access before release.

The big change

What changed: The coding assignment includes confirmation, sign-in, persistent sessions, recovery, sign-out and owner-scoped data operations in one reviewable change.

Why it matters: A signed-in screen can still expose another user's rows if the database or an existing server route trusts browser state.

What to watch: The code below is a documented reference. The owner's own callbacks, email delivery, migrations and access checks determine whether the actual app is ready.

Read the full story
—
Back to the briefing
guide

Run two Claude Code Desktop tickets in separate projects

Set up two local Claude Code Desktop worktree sessions for unrelated Git repositories, then verify each ticket's permissions, diff and tests independently.

The big change

What changed: Claude Code Desktop supports local sessions in Git worktrees, a split view, an integrated terminal and a file-by-file Changes view. The pane and terminal layout used below requires Desktop v1.2581.0 or later.

Why it matters: You can run two approved tickets in two independent repositories and assess each result against its own acceptance criteria, without making an integration branch between them.

What to watch: Git worktrees separate checkouts, while permission approvals, hooks and external resources can still reach beyond a checkout. Check the effective rules and real service configuration before either ticket begins.

Read the full story
—
Back to the briefing