Hey, it’s Marc.
“AI’s ability for “novel discovery” is starting to happen, with recent scientists across fields using the tool for breakthroughs.”
— Sam Altman, Co-founder and CEO, OpenAI
Devday 2025 for OpenAI was a spectacle of flashy launches, but the real story was Codex, now fully enterprise-ready. The AI software engineer drives 70% productivity gains internally, reviewing nearly all code and catching hundreds of bugs daily. Companies like Cisco are already embedding it, cutting code review times by 50% and accelerating projects from weeks to days. Whereas, IBM launches Project Bob, an AI partner boosting developer productivity by 45%.
The takeaway: AI is now doing the heavy lifting in software development. OpenAI is doubling down on long-running task handling, adaptive reasoning, and enterprise-grade controls to catch up with Anthropic in AI coding.
Top AI signals this week:
Anthropic launches Petri, an open-source tool for automated AI safety audits. Link
IBM introduces IBM Network Intelligence. Link
OpenAI partners with AMD to deploy 6GW of GPUs. Link
Google introduces Gemini Enterprises. Link
Adobe launches AI agents to streamline B2B marketing and sales. Link
Spotify now offers personalized music and podcasts directly in ChatGPT. Link
👉 Get your brand in front of 30,000+ decision-makers — book your ad spot now.
Top Boardroom Reads This Week
Sam Altman on Zero-Person AI Companies, Sora, AGI Breakthroughs, and more (Rundown)
OpenAI + NVIDIA: $100B Bet on 10GW AI Infrastructure (Fiftyone)
The 2025 AI Index Report (Stanford)
State of AI Report (Air Street Capital)
Power Moves
Google’s Gemini 2.5 will now run your software
Google is releasing its Gemini 2.5 Computer Use model, an AI agent that directly interacts with graphical user interfaces (GUIs). It takes a user request and a screenshot, then navigates websites and apps to complete multi-step tasks like filling out forms or scheduling appointments, no API integration needed. [RELEASE]
So what? Google’s new Gemini 2.5 Computer Use model could kill off manual “click-and-type” work forever. Early testers report huge efficiency gains. Poke.com says it’s 50% faster than competing solutions, while Google’s own payments team used it to fix 60% of UI test failures that used to take days. For enterprises, this means automation can finally reach legacy systems and complex third-party tools that never had APIs, and automation will go beyond booking, buying and other general examples.
Amazon’s new AI agent plugs into 1,000+ apps
AWS has launched Amazon Quick Suite, an agentic AI application that connects to enterprise data and applications to automate tasks, conduct research, and visualize data. It integrates with over 50 native connectors (think SharePoint, S3, Snowflake) and over 1,000 other apps via partners like Zapier and Workato. [RELEASE]
So what? Tech giants are now focusing on making AI accessible than building top models. In comparison to Google Enterprise and Microsoft Copilot, it has 33% cost advantage. It also provides more flexibility for organizations already using AWS services, eliminating the need for separate BI and automation tools.
Spotlight: AgentKit by OpenAI
A new way to build, deploy, and optimise AI agents at scale.
AgentKit is OpenAI’s end-to-end toolset for creating, deploying, and optimising AI agents. It consolidates what was previously fragmented across multiple platforms, agent orchestration, connectors, prompt tuning, evaluation, and UI embedding, into a single ecosystem. [RELEASE]
Why it matters: OpenAI aims to build a one-stop solution for developers and enterprises trying to develop production-grade AI agents. It helps them to create a unified workflow with centralised connectors, avoiding costly and slow iteration cycles.
AgentKit is built around four key components:
Agent Builder: Visual canvas to design and version multi-agent workflows
Connector Registry: Central place to manage data & tool connections across ChatGPT and API
ChatKit: Embed agentic chat interfaces directly into products with minimal dev work
Evals & RFT: Built-in evaluation and reinforcement fine-tuning to improve performance
Competitive landscape: While n8n and Make.com are powerful for general automation, AgentKit is designed specifically for AI-native, production-grade agentic workflows with deeper evaluation and governance baked in.
Use cases: AgentKit is designed for enterprise deployment at scale. Key applications include:
Customer Support: Deploy internal or external support agents with guardrails and data connectors (e.g., Dropbox, Google Drive, SharePoint).
Knowledge Assistants: Build internal copilots for teams with centralised governance.
Sales & Ops Automation: Enable sales and ops agents to integrate with CRMs and ERP systems.
Research & Intelligence: Multi-agent research workflows with trace grading and evaluation.
Product Integrations: Embed agentic chat interfaces into customer-facing products in hours, not weeks.
Bottom line: OpenAI is now focusing on AI-native workflows to target enterprises and make money providing real world applications. With AgentKit, it helps operationalize reasoning inside products, with evaluation, safety, and performance baked in.
👉 We’re tracking 1000s of AI and blockchain vendors so you can find the right vendors, partners, and bet on the right tech. Sign up for early access.
More from 51:
That’s all for today.
Thanks,
Marc & Team