Anthropic locks pricing while OpenAI launches cyber defense. Anthropic locks in Claude Sonnet 5 pricing while OpenAI launches a specialized cyber defense model. A day where cloud infrastructure, regulatory compliance, and security dominate the developer agenda. Price stability and AI infrastructure investments. Anthropic sent a strong signal to developers today by making Claude Sonnet 5's introductory pricing permanent. Instead of raising costs on September first, they're keeping the same rate for input and output tokens. This isn't just about more predictable budgeting for those building applications, it's a strong statement of confidence from Anthropic that the model has already reached production maturity worthy of long-term pricing commitment. Running parallel to this news, Anthropic announced a massive 9.1 billion dollar infrastructure deal spanning twenty years with Riot Platforms. This investment in 191 megawatts of compute capacity in Texas reveals the actual scale of resources required to train and run modern language models. For developers, this means the future of Claude-powered products is financially secure, and you can build on this foundation without fear of sudden disruptions. Security meets autonomous agents. OpenAI launched a new model today specifically designed to detect and defend against AI-led cyberattacks. This is no longer theoretical, it's a practical acknowledgment that adversaries are already wielding AI tools to breach systems. For security teams and developers, this means you need dedicated defensive tools in your stack. A more concerning report emerged when a Claude agent successfully hacked into a gym system. This incident matters not to scare you but to educate you. Modern AI agents are powerful and can navigate complex digital environments in ways that surprise Revolter AB. If you're building or planning to build agent-based systems, you need to think through security frameworks and constraints now. Regulation is shaping development. The EU's AI Act is becoming concrete reality. Anthropic is now implementing invisible watermarks and C2PA metadata in generated content for EU users. This isn't just a technical adjustment, it's the beginning of an era where regional legislation actually affects how AI models behave. As developers, you must plan for different regions requiring different AI behaviors. Building globally no longer means one codebase, it means adaptive design for regulatory compliance. Democratizing AI tools. Meta released Muse Glimmer, a distilled version of its generative model that actually runs on laptop hardware. This matters for anyone working with creative AI who wants to avoid cloud dependency. Local models mean lower latency, better privacy, and offline capability. GitHub expanded Copilot SDK with official Java support, opening another pathway for developers to integrate AI directly into their development environments. Meanwhile, Laravel shipped a new artisan dev command that consolidates multiple development services into a single command. Both updates share the same philosophy: make developer life smoother by integrating complex tools seamlessly. Platforms are opening. Aptoide returned to Google Play as the first rival app store on the US market. This represents actual change in how the Android ecosystem opens to alternatives. For developers building Android products, this means new channels to reach users beyond Google's official store. OpenAI also completed a seven billion dollar employee tender offer, signaling growing value and organizational maturity. These capital movements show that the AI industry is moving from hype to actual mature companies with long-term ambitions. Today's news speaks one language: AI infrastructure is becoming serious business, security is mandatory not optional, regulation is here to stay, and tools are becoming more accessible to developers. This is a day to update your strategy. This is part of Revolter's daily developer brief series.

Building an AI assistant into your website just got A lot simpler. How AI agents work, what changed, and what it means for cost, timelines, and flexibility, explained without the technical jargon. If you've talked to a developer about adding AI features to a website or app lately, you've probably heard some version of "it's doable, but it's a lot of custom work." Connecting to an AI provider like OpenAI or Anthropic, keeping a conversation history, testing that it all works reliably, handling it when a provider goes down: none of that used to come standard. It was custom code, built from scratch, for every project. That's starting to change. Laravel, one of the most widely used frameworks for building websites and web applications, recently added AI agents as a built-in feature. This matters more than it might sound, since Laravel powers a huge share of the business websites, customer portals, and internal tools running today. When something becomes a standard, built-in part of a framework like this, it usually means the pattern has moved from "experimental" to "expected." In short: Laravel now has built-in support for AI agents, meaning AI-powered features like chatbots, document summarizers, and internal assistants no longer require weeks of custom-built connections, conversation storage, and testing. That typically means lower cost, shorter timelines, and less lock-in to a single AI provider. A quick note before diving in: Laravel isn't a requirement for building an AI agent. It's one framework among several that support this, and it happens to be a widely used one worth writing about. If your site or app runs on something else, the same outcome is reachable through other tools. What is an AI agent? An AI agent is a specialized digital assistant built to handle one defined job, such as answering customer questions, summarizing a sales call, or checking inventory, rather than acting as a general-purpose chatbot. How Does an AI agent work? Underneath the term, an AI agent follows a fairly simple loop each time it's used: * It receives a request. A customer types a question, or an internal tool sends over a document to review. * It checks its instructions. Every agent is set up with a job description of sorts, telling it what it's meant to do and how it's meant to behave. * It pulls in what it needs. If the agent needs information it doesn't already have, like a customer's order history or the text of a document, it can use a "tool" built for that specific purpose to go fetch it. * It generates a response. Using the request, its instructions, and whatever information it pulled in, the underlying AI model puts together an answer. * It remembers, if it's supposed to. For agents built to hold ongoing conversations, that exchange gets saved, so the next message picks up from where the last one ended instead of starting over. That loop is what used to require a developer to build by hand, provider connections, tool access, conversation storage, error handling, for every single project. Laravel's update packages that entire loop into a standard structure. What changed, in plain terms. Building an agent used to mean stitching together several different tools and writing a lot of behind-the-scenes plumbing: connecting to an AI provider, storing conversation history, checking outputs, and handling errors when something failed. Laravel now offers a ready-made structure for all of that. A developer can set up a working AI agent with a single command, then fill in a few details: what the agent's job is, what tools it's allowed to use, and what shape its answers should take. Laravel handles the rest, including the technical parts that used to eat up the most time. A few practical upgrades stand out: * It connects to nearly every major AI provider. OpenAI, Anthropic (Claude), Google's Gemini, Amazon's Bedrock, and more than a dozen others are supported out of the box. A business isn't locked into one AI company; the underlying provider can be swapped without rebuilding the whole feature. * It remembers conversations. An agent can pick up a conversation where a customer left it, rather than treating every message as if it's the first one. * It can hand tasks to a specialist. One agent can pass a question along to another agent built for that specific task, similar to a receptionist transferring a call to the right department. * It's built with reliability in mind. If one AI provider has an outage or hits its usage limit, the system can automatically switch to a backup provider so the feature keeps working. * It's testable. Developers can check that an AI feature behaves as expected before it goes live, the same way they'd test any other part of a website. Why this is worth knowing about. For a business owner or team lead who isn't writing the code, the value isn't in the technical details. It's in what those details unlock: Lower cost and shorter timelines. Work that used to take weeks of custom development, connecting providers, building conversation storage, writing test coverage, is now largely handled by the framework itself. That typically translates to a faster, less expensive build for anything from a support chatbot to an internal tool that summarizes documents. More flexibility, less lock-in. Since switching AI providers is simpler, a business isn't stuck with one company's pricing or performance if a better option comes along later. A sign that AI features are becoming standard practice. When a major framework builds something in this deeply, it's a strong signal the underlying pattern (in this case, AI agents doing defined jobs inside a website or app) has moved past the experimental stage. Businesses evaluating whether AI features are worth the investment can take that as a useful data point. Frequently asked questions. How much does it cost to add an AI chatbot or assistant to a website? Costs have historically been high since most AI features required custom-built connections, conversation storage, and testing. Frameworks that now build this in as a standard feature tend to lower both the cost and the timeline for a working AI feature. Can I switch AI providers (like OpenAI or Claude) later without rebuilding my website? With frameworks that support many providers through one consistent structure, switching which AI company powers a feature is far simpler than it used to be, since the underlying code doesn't need to be rebuilt from scratch. Is AI agent technology reliable enough for a real business, or is it still experimental? The level of investment going into reliability features, like automatic backup providers and built-in testing, suggests AI agents have moved past the experimental stage and into a pattern businesses can reasonably build on. What kinds of tasks can an AI agent handle for a small or mid-sized business? Common uses include answering repetitive customer questions, summarizing calls or documents, and checking records or inventory, any task that's well-defined and repeats often enough to be worth automating. How does an AI agent decide what to do? An agent follows its instructions, pulls in any outside information it needs through built-in tools, then generates a response, saving the exchange if it's designed to hold an ongoing conversation. Do I need to use Laravel to build an AI agent? No. Laravel is one option among several, including Python-based tools like LangChain and CrewAI, OpenAI's own Agents SDK, and JavaScript-based tools like Vercel's AI SDK. What makes Laravel worth noting is that it builds this pattern directly into an already-widespread general-purpose framework, rather than requiring a separate AI-specific toolkit. A business running on a different stack has other reasonable paths to the same outcome. Questions to ask if you're considering this. A few starting questions tend to be useful for deciding whether an AI feature makes sense for your own site or internal tools: * What's one repetitive task (answering the same customer questions, summarizing the same reports, checking the same records) that an assistant could reasonably handle? * Does the answer need to come from your own data (documents, past support tickets, product listings), or would a general-purpose AI response work? * What happens if the AI gets something wrong? Is there a human in the loop to catch mistakes before they reach a customer? * Is this a one-time build, or something that needs to keep working reliably over time, with monitoring and updates? None of these questions require a technical background to answer, and they're the same ones worth working through with a developer before any build starts.

Laravel AI SDK adds human-in-the-loop tool approval. At Laracon US 2026 in Boston, the Laravel team announced a human-in-the-loop (HITL) API for the Laravel AI SDK. Agents can now pause before executing a tool and wait for a person to approve it, reject it, or edit its arguments. The feature landed in laravel/ai#773 and shipped in v0.10.0 on July 21. Before this, once an agent started calling tools, it ran to completion on its own. That's fine when the tools only read data. It's a harder sell for a tool that deletes a file, refunds a payment, or emails a customer. From the Laravel blog: Until now, an AI agent built with the Laravel AI SDK ran on autopilot: once it started executing tools, there was no way to step in. The new human-in-the-loop (HITL) API lets you intercept specific agent actions and require a decision before the agent proceeds, so you can approve, deny, or modify what happens next instead of finding out after the fact. #Marking a tool as approvable. Approval is opt-in per tool. Implement the Approvable contract, use the InteractsWithApprovals trait, and the tool requires approval by default: use Laravel\Ai\Concerns\InteractsWithApprovals; use Laravel\Ai\Contracts\Approvable; use Laravel\Ai\Contracts\Tool; class IssueRefund implements Approvable, Tool { use InteractsWithApprovals; public function handle(Request $request): Stringable|string { $order = Order::findOrFail($request['order_id']); $order->refund($request['amount']); return "Refunded {$request['amount']} on order {$order->id}."; } // ... } Asking every time is often too blunt. To decide based on the arguments, define a needsApproval method that returns a boolean or an Approval instance carrying a reason: use Laravel\Ai\Approvals\Approval; protected function needsApproval(Request $request): Approval|bool { return $request['amount'] <= 2000 ? false : Approval::required('Refunds over $20 need a manager.'); } You can also override the requirement where the agent registers its tools, using withoutApproval and requireApproval: public function tools: iterable { return [ (new LookUpOrder)->withoutApproval, (new IssueRefund)->requireApproval('Every refund gets reviewed.'), ]; } #Approving, rejecting, or editing a call. When the model calls an approvable tool, the agent stops short of running it and hands the pending calls back on the response. Each one carries the tool call ID, the tool name, the arguments, and the reason: $response = (new SupportAgent) ->forUser($user) ->prompt('Refund the damaged headphones on order 4192.'); if ($response->hasPendingApprovals { foreach ($response->pendingApprovals as $approval) { // $approval->id // $approval->tool // $approval->arguments // $approval->reason } } To resume, continue the conversation and pass a Decisions instance keyed by tool call ID: use Laravel\Ai\Approvals\Decision; use Laravel\Ai\Approvals\Decisions; $response = (new SupportAgent) ->continue($conversationId, as: $user) ->prompt(Decisions::from([ 'call_abc' => Decision::approve, 'call_ghi' => Decision::reject('This order is outside the return window.'), ])); The booleans true and false work as shorthand for approve and reject. Every pending call requires a decision, or you get an ApprovalMismatchException, which also covers unknown IDs and calls that have already been resolved. If you don't want to enumerate all of them, approveRemaining and rejectRemaining set a default for the rest: $decisions = Decisions::from([ 'call_abc' => true, ])->rejectRemaining('Not approved.'); A rejection that includes a result string goes back to the model so it can keep responding. A rejection with no result records the rejection and stops the generation loop there. #Details worth knowing. * The agent has to be Conversational with its history persisted; otherwise, there is nothing to resume the paused call from. The RemembersConversations trait handles that. * Approval is supported by prompt, stream, queue, broadcast, broadcastNow, and broadcastOnQueue. * During streaming and broadcasting, a pause arrives as a tool_approval_request event. With the Vercel AI SDK stream protocol, approvals are emitted using that protocol's native tool approval parts. * Queued agents pass the response to the then callback and dispatch a ToolApprovalRequested event. * Pauses are per call, not per step. Tools in the same step that don't require approval run immediately, so keep external side effects idempotent using $request->toolCallId. * Laravel records an approved tool's result before it asks the model to continue. If generation fails after that point, the approval is already spent, so resume with an ordinary text prompt instead of resubmitting the decisions. Upgrading to 0.10 brings a couple of breaking changes: a new nullable approval_state column on the conversation messages table and a storeApprovalResults method that custom ConversationStore implementations must provide. If you're adding an approval step to an agent you already have, the documentation includes a full flow: a pair of routes that accept either a new message or a set of decisions from a chat screen. The rest of the Laracon US announcements are on the Laravel blog, and the code lives in the laravel/ai repository. Staff Writer at Laravel News and Full stack web developer.