Linear Hits $100M ARR with Over 50% of Tickets Handled by AI Agents
Linear Hits $100M ARR with Over 50% of Tickets Handled by AI Agents
Linear COO Cristina Cordova revealed that the company crossed $100M in annual recurring revenue within months, with over 50% of tickets written by AI agents. Linear’s go-to-market strategy skips traditional sales development reps and forward-deployed engineers, relying instead on AI-driven growth. The numbers suggest AI agents are materially replacing parts of sales and support headcount—a useful reference point for SaaS growth models.
OpenAI Built a Codex-Powered 'Performance Factory' to Keep ChatGPT Fast
OpenAI detailed its Performance Factory, a system of Codex-powered workflows that automatically finds and fixes performance issues to keep ChatGPT fast. The setup turns performance tuning from manual debugging into a pipeline. For developers, it’s a concrete example of using AI code models for infrastructure automation—not just writing code, but locating and fixing production performance bottlenecks.
Gemini API Launches Tool Retrieval, Letting Agents Access Hundreds of Tools with Fewer Tokens
Google launched Tool Retrieval in preview for the Gemini API. Developers can give agents access to hundreds of tools while using fewer tokens by retrieving only the tools relevant to the current task instead of stuffing all tool definitions into context. For anyone building multi-tool agents, this cuts token costs directly and keeps agents workable as tool counts grow.
GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance
A new paper introduces GUARD to address stability in recursive self-improvement for autonomous AI agents. These agents pair a frozen base model with an editable harness of prompts, skills, tool rules, and memory, self-modifying from experience without weight updates. Without external graders, self-editing can go off the rails. GUARD uses regret guidance to constrain the process—a concrete proposal for researchers working on agent self-evolution and its failure modes.
TRACE-MAS Diagnoses and Repairs Collaboration Failures in Multi-Agent Systems
Researchers present TRACE-MAS, a framework for diagnosing and repairing collaboration failures in LLM-based multi-agent systems (MAS). Existing safeguards mainly perform agent-level anomaly detection, overlooking failures that arise from interactions such as message exchanges and task dependencies. TRACE-MAS extends diagnosis to the collaboration layer, identifying where agents fail to work together and repairing those failures. The work offers a new approach to reliability in multi-agent systems deployed for complex, multi-step tasks.
NarrowSpec: Draft Hidden Width Need Not Match the Target in Speculative Decoding
Speculative decoding speeds up LLM generation by having a small draft model propose tokens that the target model verifies in one forward pass. Draft models that read the target’s hidden states, such as DFlash, were assumed to need matching hidden width. NarrowSpec shows that the draft’s hidden width can be narrower than the target’s while still effectively leveraging target hidden states. The result relaxes a key design constraint and lowers compute overhead for this class of draft models, giving speculative decoding more flexible architectural options.
Optimizing Beam Search for Generative Recommendation Inference
Generative recommendation (GR) inference differs sharply from conventional LLM serving: long input sequences, wide beam search, and only a few fixed output tokens. Existing engines such as vLLM process beams as batched independent sequences, which is inefficient for this workload. The researchers propose optimizing beam search first and kernels second, redesigning beam scheduling around GR’s specific load characteristics. The approach can improve throughput and resource utilization for online GR inference, offering practical guidance for teams deploying large-scale recommendation systems.
GRAFT: Execution-Guided Genetic Recombination for Agent Design
Automated agent design searches for executable architectures that improve LLM-based agent performance, with evolutionary methods using language-model crossover to recombine mechanisms for task decomposition, model coordination, and verification. GRAFT introduces execution-guided genetic recombination, using actual execution results to steer the recombination process and more effectively select promising agent architectures. The method points to gains in both search efficiency and final performance, relevant to developers customizing agent pipelines for specific tasks.
QDA: Closed-Loop LLM Quantization Design with Agents
Post-training quantization is the standard way to serve large language models at low cost, but the best recipe varies across models and precisions, forcing experts to redesign it for every new model. QDA builds a closed-loop quantization design process with agents that automatically searches suitable recipes for different models and precisions. The approach reduces repetitive manual tuning and could let quantized deployment adapt to new models faster, with practical value for engineering teams managing inference costs.
PostgreSQL 19 Ships Two Months Late After AI-Fueled Vulnerability Backlog
PostgreSQL 19 is out. The release normally lands in September but slipped two months this year, mainly because the team had to work through a surge of AI-reported vulnerabilities. Simon Willison says the release was worth the wait. For Postgres users, specific features are in the official release notes—but the delay itself shows how AI code-auditing tools are affecting open-source maintenance cadence.
Gemini 4 Neon Scores 61 on Intelligence Index; Anthropic Prices IPO at $83
Google’s mid-tier Gemini 4 Neon scored 61 on the Intelligence Index, positioned as a contender against Anthropic’s Opus/Sol line. Separately, Anthropic priced its IPO at $83 per share. The two items reflect model capability competition and public-market valuation of AI companies. For industry watchers, a mid-tier model approaching flagship scores signals intensifying competition in the cost-performance segment.
Vercel CLI Adds Billing Commands for Invoices and Budgets
Vercel CLI now supports billing management. Users can check their team’s billing overview, review and download invoices, and preview upcoming charges directly from the command line. All billing commands, including usage, contracts, and budgets, are grouped under ‘vercel billing.’ Billing overview and invoice history are available on all plans. For dev teams, routine billing checks no longer require opening the web dashboard—everything can be done in the terminal.
Walter Isaacson on Curiosity and Creativity in the AI Age
Biographer Walter Isaacson joined Google for a conversation on how history’s greatest innovators can inspire creativity and human judgment in the AI age. The episode explores curiosity, interdisciplinary thinking, and a healthy skepticism toward technology, arguing that individual judgment and original thinking remain irreplaceable as AI tools become more widespread. The episode is available on the Google AI Blog for audiences interested in the intersection of AI and the humanities.
GitHub Launches Replica and Open Source Accessibility Framework
GitHub introduced two new open source projects: Replica and the Open Source Accessibility Framework, aimed at collaboration and inclusion. Specific capabilities are in the project docs. For open source maintainers, the accessibility framework offers a reusable set of guidelines that lowers the barrier to making projects usable for people with disabilities.
OpenAI Codex Releases rust-v0.171.0
OpenAI Codex shipped rust-v0.171.0, alongside pre-release updates 0.171.0-alpha.13.1 and 0.172.0-alpha.2. This is a routine version bump for the Codex Rust component; change details are in the release notes. Developers using Codex can upgrade to 0.171.0 for the latest fixes.