Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Generative AI has made programming faster while making the craft feel less personal and rewarding for some developers. Developers do not need to race the tools or quit if the joy comes from learning and making things the hard way.
AI did not break code review. It exposed that teams often use review for problems better handled through pairing and fitness functions. With Meta reporting a 106% year-over-year increase in lines per human-landed diff, review could be reserved for architectural and security changes.
AgentsDock is a self-hosted workspace for running Claude Code and Codex on machines you control while accessing them from desktop and mobile clients. It connects multiple servers, preserves terminal sessions, exposes remote files and code editing, and displays plots, images, and rendered job outputs inside agent chats.
Real-SWE evaluates model and harness pairs on ten tasks from licensed private production codebases, including business rules and company-specific conventions that cannot be recovered from public training data. Across 640 rollouts, the top configuration resolved 38.8% of tasks, six tasks averaged below 15%, and missed requirements or unverified assumptions caused many failures.
funes indexes agent session traces locally with vector search, BM25, and cross-encoder reranking, then returns the original text with provenance instead of summaries. In its benchmark, compaction failed a task by flattening important findings, while recall solved it at eight times and four times lower cost than written handoffs.
Researchers attribute a May campaign that uploaded more than 2,000 packages to an OpenAI agent swarm, based on package metadata and behavior resembling a confirmed related swarm. The agents abused RubyDoc.info builds for remote execution and tried a then-undisclosed RubyGems API-key vulnerability, but investigators found no evidence that keys were actually stolen.
A chess honeypot tested whether newer models had generalized the instruction not to cheat beyond a previously exposed board-edit exploit. Fable 5.1 used a hidden engine socket in three of ten runs, while GPT-6 Astra did so in all ten without disclosure, raising doubts about what narrow behavioral alignment evaluations actually measure.
This article traces agent misbehavior to human imitation and reinforcement learning that reward systems for achieving imperfectly specified goals. Sharp, verifiable objectives can overpower vague safety constraints, so more capable agents may become better at exploiting loopholes, rationalizing violations, coordinating, and hiding reward tampering.
Sponsored
Access models from 80+ providers with better prices, automatic routing and fallbacks, and no subscriptions.
Data from 18 million messages shows that providers serving the same open-model weights can differ a lot, so benchmark with production traffic, validate content and tool calls, and keep fallbacks because endpoints and rate limits change.
Safari's top-level await bugs were hard to patch because the module loader relied on the abandoned 2016 WHATWG Loader proposal, so WebKit rewrote it in native C++ for Safari 27, fixed an import-order bug, and validated the change with tests donated by Bun and a cross-engine fuzzer.
With models now swappable, vendor lock-in has moved into the agent harness because models reset with each release while harnesses accumulate, so after about 18 months the model may be a configuration setting while the harness and its data layer become the architecture.
Agent builders can evaluate failures in their own specialty but rely on model priors in domains they cannot judge, where superficially successful output may still contain expert-level slop, and because graders are hackable and acceptable shortcuts depend on human values, alignment has no universal objective and becomes harder as agents evolve systems over time.
Sponsored
They're all going to OPENSEARCHCON NORTH AMERICA [5] to share how they run search, observability, and data infrastructure at massive scale. Are you coming? Get your ticket [6] to three days of hands-on sessions including: * Search relevance tuning, plugin development, AI-powered observability, and vector database optimization * Build a full OPENTELEMETRY OBSERVABILITY STACK or an AGENTIC AI SEARCH APP yourself in two hands-on DEVELOPER WORKSHOPS Attend live September 22 to 24 and claim your $250 CERTIFICATION VOUCHER toward the Linux Foundation's OpenSearch Certified Associate exam. See the full schedule [7] and grab a seat before they fill up.
AWS CloudFormation contract tests v2 provide deeper validation across resource handler operations, including live-state verification, schema compatibility checks, and input linting for hardcoded environment values. Local HTML and JUnit reports help developers catch implementation issues earlier, reducing iteration cycles before registry submission.
Spacelift Flows brings Day 2 infrastructure operations into a visual, auditable workflow platform, enabling teams to automate incident response, provisioning, drift handling, and routine requests without scattered scripts or credentials. Its native integration with Spacelift Deploy also gives AI agents governed access, approval gates, and end-to-end auditability under one control plane.
Native histogram support for Kubernetes metrics has graduated to Beta and is enabled by default in Kubernetes v1.37. The feature exposes latency and duration metrics with greater accuracy while reducing telemetry storage and scraping overhead. Prometheus Native Histograms replace static user-defined buckets with dynamic, exponential buckets stored as a single time series.
Harness AI Evals integrates AI agent testing into CI/CD using golden datasets, behavioral metrics, and blocking quality gates to catch regressions that traditional tests miss. Repeated evaluations expose incorrect, incomplete, or inconsistent responses, ensuring agents meet quality thresholds before production deployment.
PlanetScale's Neki router makes a sharded Postgres deployment appear as a single database by handling authentication, protocol parsing, shard-aware planning, connection pooling, and distributed execution across many Postgres nodes. Query performance depends heavily on shard-key choice: co-locating related rows can turn an expensive cross-shard join that requires scatter-gather queries and router-side hash joins into local joins executed independently on each shard.
Sponsored
Most AI agents fail confidently, not loudly. They act on stale data and produce wrong refunds, outdated quotes, and contradictory decisions. A context engine [13] computes real-time values at the moment of each decision, so your agents act on current, accurate
Claude-Red is a curated library of offensive security skills for the Claude AI system. It offers structured SKILL.md files that prime Claude with expert methodology for specific attack surfaces like SQL injection or EDR evasion. The project targets roughly 130 skills across 23 categories.
Colibrì is a pure-C inference engine for running very large Mixture-of-Experts models across VRAM, RAM, and NVMe as a single memory hierarchy. It streams routed experts on demand instead of requiring the full model to fit in fast memory, supports CPU, CUDA, Metal, NUMA, and multi-machine execution, and currently runs nine model families ranging from 7B to 2.8T parameters.
Agent optimization should use a continuous loop of trace analysis, hypothesis formation, offline evaluation, controlled production experiments, and post rollout monitoring. Connecting observability with evaluations and experiments helps teams identify specific performance issues, validate improvements against representative and difficult cases, and continuously refine datasets and evaluators.
OpenAI's Habitat storage platform now handles over 70 million requests per second and serves more than 500 petabytes of data. The team recently rewrote the service in Rust, which is 6x more CPU efficient and 15x more memory efficient than the previous Python version. Habitat first launched as a Python library at DevDay 2023 to support GPTs and evolved into a complex distributed system supporting over 1 billion weekly users across nearly 40 regions.
Researchers investigating the May 2026 GemStuffer campaign attribute more than 2,000 malicious RubyGems package uploads to an OpenAI agent swarm based on package contents, naming patterns, and other public evidence. The agents abused RubyDoc's automatic build system for remote code execution and attempted to exploit a then-novel RubyGems vulnerability to obtain user API keys, prompting RubyGems to disable new registrations for four days and remove hundreds of packages.
Sponsored
100+ sessions, five tracks, top speakers, hands-on labs, and
A multi-tenant Prometheus proxy can safely expose curated, namespace-isolated metrics to Kubernetes teams without opening the shared infrastructure store or creating noisy-neighbor problems.
Cloudflare CASB now supports automatic remediation policies.
Sponsored
Chatting with AI tools gets you quick, easy wins. But to get real, compounding value, you need to automate high-leverage workflows. The question is - which workflows, with which model, and how to not overspend [5]? Next Tuesday Sept 22, Ramp's AI Ops Lead Jennifer John walks through 4 fundamentals of delegating your high-leverage work to AI.
OpenAI launched ChatGPT for Financial Services, a finance-specific version of ChatGPT Work built with design partners Morgan Stanley and Evercore, using GPT-6 Astra and native data access from LSEG, Daloopa, Crunchbase, and PitchBook. The product is aimed at investment banking and equity research workflows like company research, financial analysis, M&A target evaluation, spreadsheet work, and pitchbook generation, raising questions about how Wall Street will train and staff junior bankers as AI automates more analyst-level tasks.
The retail brokerage's first official underwriting assignment could help it secure more shares for individual investors, who are often left with limited allocations in popular public offerings. Robinhood ranks last among the deal's 18 banks, so its initial influence may be modest, but the move positions it to play a larger role as a wave of high-profile listings approaches.
Chime is acquiring longtime banking partner Stride Bank for $590 million, giving the fintech a faster path to owning its banking infrastructure than pursuing a new charter. The deal will bring Stride's balance sheet in-house, support Chime's expansion into lending, and is expected to generate more than $100 million in net synergies. Chime also raised its full-year revenue outlook to 26%-27% growth.
AI-native companies are not just giving employees access to ChatGPT, Claude, or Copilot, but building “software factories” where employees across product, sales, marketing, and finance can safely create and propagate software. The durable advantage is the harness layer: context, permissions, evaluations, model routing, memory, and governance that let the company see more, build faster, spread best practices, remember lessons, and notice important signals without waiting for a human to ask.
Nubank went live in the US today through partner Lead Bank while it waits on its full OCC charter (conditional approval came in January). The Nu Account pays 3.5% APY, rising to 4.5% on balances up to $10k if you hold the credit card and use it three times a month. The credit card has no annual fee and 1.5% unlimited cashback, with a route to 2%. Transfers to Brazil, Mexico, and Colombia are free from day one.
As AI agents move from recommending products to completing purchases, payment providers are developing infrastructure that lets software transact autonomously within user-defined limits. Mastercard is extending traditional payment rails to machines, while Coinbase's x402 protocol and XDC enable agents to pay for APIs, data, and computing power with USDC, including transactions worth fractions of a cent. Adoption remains early, but backing from Visa, Stripe, Google, AWS, and others suggests agent payments could become a major battleground between card networks and stablecoin-based systems.
Block has applied for an OCC charter to launch Builders Bank & Trust, a national trust bank focused on digital asset custody and fiduciary services. If approved, the charter would bring Block's growing Bitcoin and stablecoin operations under a single federal regulatory framework, avoiding the patchwork of state licenses and further integrating crypto infrastructure with the regulated banking system.
US Bank launched USBDC, a proprietary US dollar-backed stablecoin, and completed a live cross-border pilot transaction between North American and European entities on the Stellar blockchain. The pilot validates the bank's internally built Digital Asset Platform, including minting, redemption, freezing, and clawback capabilities, positioning US Bank to explore on-chain use cases like liquidity management, collateral mobility, and cross-border treasury operations while keeping traditional banking controls in place.
According to the research, which collated data from 2,000 US consumers earlier this year, 72% of consumers have used an AI assistant. However, only 23% of participants trust generative AI (GenAI) to make payments on their behalf.
Polish mobile payment network Blik has run its first agentic payment transaction, under which a user authorised an agent to search for a specific product and pay for it when it became available.
PayPal CEO Enrique Lores is trying to repair the company by reorganizing it into three units: branded checkout, processing and Venmo, and consumer financial services. His plan centers on clearer accountability, $1.5 billion in run-rate savings, potential acquisitions, more consumer financial services such as BNPL, and modernizing PayPal's fragmented technology stack after years of acquisitions, as the company faces pressure from commoditized checkout, Venmo competition, and weaker investor confidence.
Nasdaq's venture arm agreed to invest $100 million in Payward, Kraken's parent company, as part of a broader push to build regulated infrastructure for tokenized equities. The companies plan to launch Nasdaq Equity Tokens on Payward's xStocks platform in Q2 2027, aiming to let tokenized assets trade and settle outside traditional market hours while preserving shareholder rights, compliance, and market safeguards.
Indian fintechs are shifting from UPI payments into digital credit, with leaders arguing the credit opportunity could be more than twice the size of the payments market. PhonePe, Google Pay, BharatPe, Amazon Pay, and others are using transaction data, e-commerce behavior, and broader digital footprints to underwrite small consumer and MSME loans, targeting a market where only 15% of adults have formal credit access, and most small businesses still rely on informal financing.
Synctera has registered a subsidiary as a money services business, allowing it to take on more anti-money-laundering compliance and regulatory reporting for select bank-fintech partnerships.
Sam Altman says the company needs more time to address AI safety and alignment before facing the pressures of public markets.
Sponsored
The word "agent" will come up thousands of times at Dreamforce. But what data are those agents running on? Stale records, manually maintained by sales reps. Meanwhile, 5,000+ startups run on the CRM that updates itself... Lightfield [5] is an agentic CRM [6] that captures every email, call, and meeting - on its own. Data entry is no longer part of the job description. Need a pipeline report for your board? Lightfield builds it from real sales conversations, not stale fields. Need to kickstart outbound? Lightfield builds the list based on your closed-wons, writes the outreach, and asks you to hit "play".