Date: {{current_date_full_with_day}}

Hey {{first_name | AI Visionaries}},

In this 103rd edition we dive into new releases from Open AI, Perplexity and Anthropic.

The pace of change in AI is mind-numbing and we decided to parse through the releases and brought the most relevant ones for you!

Thank you for being an engaged reader and helping us cross 5,500 subscribers over 100 editions! As usual, please send in your feedback and suggestions - it helps me to improve.

-Renjit

PS: If you want to unleash the power of OpenClaw and Hermes AI agents for personal and business productivity, you can setup time speak to me here»

OpenAI has introduced GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper, a new set of models aimed at live voice agents, translation, and streaming transcription. The move matters because voice AI is shifting from a call-and-response demo into a workflow layer that can reason, use tools, handle interruptions, and complete tasks while a conversation is still happening.

The older pattern for voice bots was simple: listen, transcribe, answer, repeat. The new direction is more operational. A service agent can understand a request, check data, trigger a tool, and keep the conversation moving without forcing the user through rigid menus.

The clearest use cases are customer support, sales qualification, travel and real estate assistance, coaching, accessibility, and multilingual service. For small teams, this could reduce the gap between a polished human concierge and a basic chatbot.

Voice is becoming a serious interface for software. Founders should start mapping which customer conversations are repetitive, high-value, and safe enough to hand to an agent first.

The opportunity is not just better voice chat; it is voice agents that can complete useful business tasks while customers are still talking.

Read the source

OpenAI has launched a Codex Chrome extension that lets Codex work inside browser sessions, inspect context across tabs, and help with web-app tasks without taking over the whole browser. It is a practical step toward coding agents that can see the same product surface a user sees.

Many product problems do not live only in the codebase. They show up in a login flow, a broken dashboard, a confusing pricing page, or a browser-only admin tool. Bringing Codex into Chrome makes the browser a workspace for debugging, testing, and research rather than a separate place the developer has to describe from memory.

Indie hackers can use this kind of workflow to inspect front-end behavior, test web apps, review authenticated tools, and work through browser-heavy tasks faster. Teams building SaaS products should also see the direction clearly: agents are moving from editing files to operating across the product environment.

Read the source

Claude's Microsoft 365 connector now lets users search and analyze information across SharePoint, OneDrive, Outlook, Teams Calendar, and Teams Chat. That puts the assistant closer to the places where business context already sits: documents, email, meeting records, and team conversations.

This is less about a new chatbot feature and more about distribution. AI adoption has often failed when teams had to leave their normal tools and paste context into a separate assistant. Connectors reduce that friction by letting the model work with existing office content.

For business leaders, the practical use cases include preparing for meetings, searching internal documents, reviewing past discussions, and summarizing scattered project context. The value is strongest when the assistant can answer with the company's own information, not just generic web knowledge.

The governance question also becomes more important. If AI is connected to documents and chats, permissioning and data hygiene need to be treated as part of the rollout, not as an afterthought.

AI assistants are becoming more useful when they work inside the office systems teams already use every day.

Read the source

OpenRouter has highlighted a broader product push that includes video generation through one API, project Workspaces, new model access, and a TypeScript SDK for agent development. The pattern is clear: AI builders want fewer scattered accounts and more control from one operating layer.

As teams use more models, the work becomes harder to manage. A founder may need one provider for text, another for coding, another for video, and separate controls for billing, projects, and usage. OpenRouter's direction is to bundle routing, access, governance, and multimodal generation into a single layer.

This matters for agencies, AI product studios, and teams running several client or internal projects. Workspaces can help separate projects, budgets, and experiments. A unified API can also make it easier to swap models without rebuilding the whole stack.

The broader lesson is that model choice is becoming infrastructure. The winners may be the teams that can test, route, and govern models without slowing down product development.

Read the source

Anthropic has reportedly secured access to SpaceX's Colossus 1 compute capacity, with coverage pointing to more than 300 MW of compute and higher Claude Code and API limits. The headline is not just about one partnership. It shows how much the AI market now depends on raw infrastructure.

For customers, compute shows up in practical ways: rate limits, latency, reliability, pricing, and whether an AI vendor can support enterprise-scale usage. A model may be strong, but if capacity is constrained, teams feel it in daily workflows.

Founders choosing AI vendors should look beyond benchmark charts. They should ask whether a provider has enough capacity to support their product, support usage spikes, and keep API access predictable. That matters especially for companies building AI into customer-facing products.

Compute is becoming a competitive moat. It is also becoming a procurement question for any business that depends on frontier models.

Read the source

Perplexity's Personal Computer brings an agentic assistant onto the Mac, with the ability to work across local files, native apps, the web, and Perplexity's server-side compute. It points to a larger change in AI products: assistants are moving from answer boxes toward direct execution on the user's machine.

A large part of business work still happens outside the browser. Files sit in folders. Notes live in apps. Research moves between tabs, documents, and email. A local agent that can navigate that environment could become more useful than a search-only assistant.

For founders and operators, the first useful jobs are likely to be admin, research, file organization, and repeatable workflows. Think of preparing a briefing folder, comparing source documents, or moving information between tools without a long series of manual steps.

This category will still need careful trust and permission controls. But the direction is important: personal AI is becoming less like a website and more like a work layer on the computer itself.

Read the source