Search the archive

WHEN THE HARD PART STOPS BEING HARD

In a world where producing work is easy, we need to be more ambitious and take on challenges we could not have imagined a couple of years ago.

QWEN 3.8 27B IS EXCELLENT, BUT IT DEFAULTS TO WILDLY OVERTHINKING THINGS

Qwen 3.8 27B is an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba's Qwen research lab.

OPENAI SHEDS SENIOR EXECS IN PRE-IPO REFRESH

OpenAI has seen significant leadership changes, including the departure of top executives like the chief revenue officer, COO, and head of ethics, ahead of an anticipated IPO.

THE AI ENGINEERING SKILLS MAP

AI engineering now centers on four skills: building and deploying AI applications, software fundamentals, coding-agent fluency, and shaping what gets built.

GPT-5.6 SOL ULTRAFAST

OpenAI previewed Ultrafast, a GPT-5.6 Sol mode capable of generating up to 750 output tokens per second and operating at up to 14x standard processing speed. The approach aimed to deliver real-time performance without switching to a smaller model.

GOOGLE'S GEMINI 3.7 FLASH TARGETS CODING AND AGENTS WITH A 50% INTRODUCTORY PRICE CUT

Google is rolling out Gemini 3.7 Flash, a new version of its workhorse AI model. The release comes just three weeks after the release of Gemini 3.6 Flash, an unusually short turnaround attributed to developer feedback and algorithmic improvements. API prices have been temporarily cut in half. Gemini 3.7 Flash will cost $0.75 per million input tokens and $3.75 per million output tokens until the end of the year.

ANTHROPIC COULD BE WORTH $2 TRILLION WHEN IT GOES PUBLIC

Investors anticipate Anthropic's IPO could exceed a $2 trillion valuation, potentially becoming the largest in history, fueled by rapid revenue growth and demand for its AI models. Despite regulatory hurdles and competition, the startup projects annualized revenue of $100-120 billion by the end of 2026. Investors remain optimistic, citing Anthropic's strong market position and performance compared to rivals like OpenAI and Google.

WILL FINANCING BOTTLENECK AI COMPUTE? AN ANTHROPIC CASE STUDY

Much of the financing Anthropic needed for its infrastructure buildout was assembled before the company's revenue spiked. The buildout suggests that financing is unlikely to be the immediate limit on frontier computer growth. Institutional investors appear willing to lend capital against commitment to long-term payments, especially if more established organizations are willing to back up part of the risk.

HOW AI AGENTS COULD FAIL AT SCALE

Anthropic examined how individually benign behaviors in frontier AI agents could compound into systemic failures when many agents interact in shared environments. The research highlighted risks including confabulation, reward hacking, and unexpected dynamics emerging faster than human institutions can oversee them.

SUBAGENTS ON SUBAGENTS: HOW MANY LAYERS DEEP IS TOO MANY?

Recursive agent systems should be designed as dependency graphs because upstream mistakes can propagate across many workers while leaf errors stay localized. Reliability depends less on nesting depth than blast radius, requiring provenance, verification, and stronger controls at high-impact nodes.

Autonomy, and Assume Misalignment. Read the paper →

AGENT PLUGINS ARE THE FUTURE OF AGENT SKILLS

Agent Plugins package skills and MCP dependencies into one portable, vendor-neutral folder that compatible clients can load. The standard reduces fragmented setup through shared manifests, portable paths, isolated failure boundaries, and client-specific extensions, while leaving authentication unresolved.

CLOUD AGENTS START 3X FASTER WITH BUILDS

Agents are only as capable as the environments they work in. Cursor can now continuously prepare development environments in the background at no additional cost. Agents can now start in ready environments, resulting in up to 3x faster responses. Agents will keep using the last successful build, and work continues uninterrupted while developers debug in the background.

OCR 4.1

Mistral OCR 4.1 is a highly specialized vision-multimodal model engineered specifically to ingest, parse, and structure complex document formats. It excels at reading complex tabular layouts, understanding hierarchical document structures, and directly outputting clean, machine-readable JSON or Markdown. The release pressures cloud providers and proprietary players to re-evaluate their pricing tiers for multimodal inputs.

GTM ENGINEER, APPLIED AI AT TLDR ($175-205K BASE + $40-60K BONUS, FULLY REMOTE)

TLDR is hiring a GTM Engineer to join our Applied AI team and own our AI-native GTM stack. We're looking for someone comfortable building AI agents and working with HubSpot. Click here to learn more [20]!

GOOGLE LAUNCHES SHEETS CANVAS FOR GEMINI MINI-APPS

Google Sheets canvas is a Gemini-powered feature that turns spreadsheet data into custom, interactive mini-apps inside Google Sheets. The new layer sits directly on top of a spreadsheet to give users a visual way to organize, edit, and navigate information without formulas, programming, or a separate app. The visual layer is not static and will update along with the spreadsheet underneath it. Sheets canvas is now available globally in English to Google AI Pro and Ultra subscribers.

OPENAI REVENUE CHIEF DENISE DRESSER LEAVING, SECOND MAJOR EXECUTIVE DEPARTURE IN DAYS

Denise Dresser, OpenAI's Chief Revenue Officer, leaves the company just ahead of its anticipated IPO, marking the second executive departure this week. Dresser, who joined OpenAI from Salesforce less than a year ago, will be replaced by Dali Rajic from Wiz. Earlier this week, OpenAI executive Brad Lightcap also announced his departure to start a new venture.

workshops — so you know where AI pays off. Take the tour

APPLE TRAINS OWN AI MODEL FOR CHINA WITH ALIBABA SUPPORT, REUTERS REPORTS

A China-trained model could give Apple greater control over its AI offerings in the country.

WRITER INTRODUCES NEW AI MODEL AND UPGRADED HARNESS TO CONTAIN TOKEN COSTS

Writer, which offers AI tools and agents for marketers, has launched a new flagship model called Palmyra X6 that provides deployment-ready capabilities at a much lower price.

GOOGLE TESTS AGENT MANAGEMENT UI ON AI STUDIO

Google is testing a new Agent management UI in AI Studio, adding a dedicated tab for managing Cloud Agents within Google Cloud projects, not in consumer-grade sandboxes.

CLAUDE'S CHROME SIDE PANEL BECOMES A FULL COWORK SESSION

Anthropic upgraded Claude in Chrome so the side panel now runs a full Claude Cowork session. Conversations save to your Claude account and resume on desktop, web, or mobile, and your existing skills and connectors work in the browser without setup.

INTRODUCING GROK 4.6

Grok 4.6 focuses on long-running agent tasks, matching GPT-5.6 Sol's performance on the Artificial Analysis Intelligence Index. It excels in turning product ideas into working versions and improving safety and capabilities for tasks like vulnerability patching and AI research. Available now in Cursor, Grok Build, and via API, Grok 4.6 offers 2x included usage for the first week.

DEEPSEEK PRICES ITS NEW V4-PRO-0813 MODEL AT $0.87 PER 1 MILLION OUTPUT TOKENS

DeepSeek-V4-Pro-0813 is now rolling out on the DeepSeek API and DeepSeek chat. The AI lab has priced the model at $0.435 per 1 million tokens of input and $0.87 per 1 million tokens of output. The model outcompetes Opus 4.8 on Terminal Bench 2.1, Cybergym, DeepSWE, and AutomationBench benchmarks. DeepSeek was second only to Anthropic in terms of the total number of tokens consumed in July.

QWEN3.8-2.4T-A95B

Qwen3.8, based on Qwen3.5's architecture, introduces advanced capabilities in coding and long-horizon tasks with improved agent execution for reliable task completion. It supports various deployment frameworks like SGLang and vLLM, offering robust integration with popular tools. The model's reasoning depth adjusts through reasoning_effort settings, enhancing performance in complex tasks.

NVIDIA IS SPEEDRUNNING THE CREATION OF A SYNTHETIC HYPERSCALER

A hyperscaler is a scaled infrastructure collective of CPUs, networking, and storage with a development platform on top. It pools and smooths the financial obligations of its users, and builds software that makes consumption of the underlying primitives simple by abstracting them away. Nvidia has built software and infrastructure that make it easy to install and run its hardware and also financing products to smooth utilization across a distributed fleet. The complaint that these platforms keep capital tethered to Nvidia and away from other ASICs/accelerators is a moat dressed up as a risk.

ENTERPRISE AI SHIFTS TOWARD EXECUTION

OpenAI published two studies showing enterprise AI use moving from assistance toward more delegated, agentic work. The highest-usage firms generated many times more output tokens per active user than typical firms and adopted connected tools and workflows more frequently.

BUILDING SAFER MCP SERVERS

This post outlines several ways to expose PostgreSQL through MCP, ranging from flexible agent-generated SQL to tightly constrained, typed query tools. The tradeoff is between agent flexibility and limiting access to only permitted database operations.

on it in production? After testing across eight dimensions

critical to enterprise MCP reliability (e.g., OAuth token lifecycle, large dataset behavior), CData found meaningful gaps. Expert guidance

helped, but not always. See where it broke

SPECULA: SCALING FORMAL SPECIFICATIONS FOR AUTONOMOUS MODEL CHECKING OF SYSTEM CODE

Specula is an agentic system that automates the process of software bug finding through authoring and model-checking a spec for the code. It derives TLA+ specifications automatically from the code, checks code-spec conformance through trace validation, model checks the spec to find concurrency bugs, and reproduces the bug at the code layer by writing integration tests with precise timing. This post looks at what Specula gets right, its major contributions, and unresolved questions about the terrain. Specula is a great pragmatic idea, and it works for what it does, but it still skirts the real hard problem of composition, so it cannot say anything about whether per-module guarantees add up to a system-level guarantee.

MICROSOFT LAUNCHES MAI-THINKING-1

Microsoft MAI-Thinking-1 is a medium-sized reasoning model aimed at cost-efficient enterprise workloads across coding, math, and knowledge tasks.

MAI-IMAGE-2.6 REACHES NO. 2 ON ARENA

Microsoft's MAI-Image-2.6 reached second place on the Arena text-to-image leaderboard.

HIRING AGENTS IS THE EASY PART

Agent adoption will be constrained less by capability than by verification: companies need systems that define quality, evaluate ongoing performance, and compound feedback. The hardest problems are tacit standards, company-specific evals, feedback ownership, permissions, liability, and self-improving learning loops.

GROK 4.6 – A FIELD GUIDE

Grok 4.6 stands out less for a single capability jump than for speed, dense communication, stronger polish, and reliable work across coding and knowledge tasks. The highest-leverage prompting pattern is short instructions plus explicit acceptance criteria and repeated self-verification.

and recorded expert sessions. Download free here

VIBE-CODING STARTUP LOVABLE HITS $13 BILLION VALUATION

The startup is on track to generate a revenue run rate of close to $600 million by the end of this month.

AS AI SAFETY CONCERNS MOUNT, THREE PIONEERS MAKE THE CASE FOR STAYING OPEN

AI researchers Geoffrey Hinton, Fei-Fei Li, and Andrew Ng advocate for keeping AI open to prevent a few large firms from monopolizing advancements.

HOW A THREE-PERSON TEAM SHIPS HUNDREDS OF PRS

This post describes Kenn Software's agent-assisted engineering workflow, where three developers merge hundreds of pull requests per week while maintaining a low production bug rate.

WHAT SORT OF MATHS ARE LLMS GOOD AT?

OpenAI's recent math announcement is extraordinarily impressive, but LLMs aren't better yet than all humans at all aspects of mathematics - if they were, then there would be much more of a flood of results.

Register

BRAD LIGHTCAP, OPENAI'S LONGTIME COO, IS LEAVING TO ‘START SOMETHING NEW'

Brad Lightcap, OpenAI's chief operating officer, is leaving to start a new venture. There are scant details about what this may involve, but Lightcap has hinted that it involves advancing OpenAI's mission from a different vantage point. OpenAI has been undergoing a shakeup of its top-level executives as it prepares for its IPO. Fidji Simo, Bill Peebles, and Kevin Weil have also recently left the firm.

GEMINI PASSED 1 BILLION MONTHLY USERS

Google said the Gemini app surpassed 1 billion monthly active users, becoming its 14th product to reach the milestone. The company also reported heavy voice usage, more than 150 million images generated daily, and over 100 million active Gemini users on iOS.