This article walks through how to use KEDA to scale Kubernetes worker pods based on Amazon SQS queue depth rather than CPU or memory usage. The core idea is that in queue-driven systems, the number of messages waiting is a more accurate signal of demand than infrastructure utilization. KEDA calculates desired replica counts using the formula: outstanding messages divided by a configured per-pod target. It can scale deployments all the way down to zero when queues are empty.
Once models are capable enough for everyday coding, research, and analysis, inference speed can matter more than marginal gains in intelligence, with roughly 100–200 output tokens per second feeling fast enough for interactive work. Further model acceleration eventually produces diminishing returns because tool calls, databases, local hardware, and human review become the dominant bottlenecks, shifting competition toward serving speed and price.
DwarfStar is a native inference engine built specifically for running DeepSeek V4 Flash and DeepSeek V4 PRO locally, with additional support for GLM 5.2. It is not a general GGUF runner. The project includes model loading, prompt rendering, tool calls, KV state, an HTTP server, and a coding agent built and tested together. It supports Metal on macOS, CUDA, and ROCm backends, and can split large models across multiple machines using pipeline or tensor parallelism over Thunderbolt or TCP.
Kaneo is an open source project management tool built around a minimal feature set, on the premise that most competing platforms include too many features rather than too few. It supports Docker Compose and Kubernetes deployments, and a CLI tool called drim handles HTTPS, database setup, and service configuration automatically.
AWS CloudFormation Express mode reports stack completion as soon as resource configuration is applied rather than waiting for full stabilization, speeding development iteration and AI agent workflows while preserving dependency ordering and rollback options.
Rust 1.98 adds algebraic floating-point operations that let developers explicitly permit reassociation and fused instructions without enabling every risky optimization associated with fast-math. Allowing the compiler to reorder independent calculations can unlock SIMD vectorization and significantly accelerate numerical workloads, with the tradeoff that results may vary slightly because floating-point addition is not associative.
Sponsored
This AWS + New Relic webinar shows you to autonomously triage incidents, pinpoint root cause, and remediate without custom tooling.
Oracle Log Analytics enables security teams to collect OCI Compute instance logs via the Add Data workflow and analyze access activity, failed logins, and privilege escalation attempts through curated Compute dashboards for deeper investigation.
Microsoft is deprecating the Azure DevOps issuer for workload identity federation service connections, retiring it by July 1, 2027, and standardizing on the Microsoft Entra issuer, excluding multitenant apps and non-public cloud scenarios.
2026.08.03·devblogs.microsoft.com·5 min read·TLDR DevOps Docker now supports OpenID Connect (OIDC) for GitHub Actions, letting workflows authenticate with short-lived, per-run tokens instead of stored personal access tokens or organization access tokens.
Sponsored
Your global expansion is profitable…for your bank. Traditional banks can drain 3%+ from every cross-border transaction through layered FX markups, SWIFT fees, and cross-border charges.
Surprised? So are many of your peers. Airwallex [5] partnered with Sapio Research to survey 1,600 finance leaders across 12 markets to uncover the operational bottlenecks holding back international growth.
Inside, you'll learn how top finance teams [6]:
* Bypass traditional banking markups by using local rails in 120+ countries. * Cut reconciliation time in half with real-time ERP syncs. * Build the clean, real-time data foundation needed to power AI and automation.
New York is suing Kalshi, alleging its sports, political, and other event contracts constitute unlicensed gambling and seeking a permanent injunction, user restitution, and penalties that could total $36 billion. Kalshi and the CFTC argue the federally licensed exchange falls under exclusive federal oversight, escalating the broader fight over whether prediction markets are financial derivatives or state-regulated gambling.
Chime is cutting about 10% of its workforce, or roughly 150 employees, as it restructures around AI and aims to improve profitability as a public fintech. CEO Chris Britt said smaller, flatter teams are enabling faster execution, reflecting a broader shift across financial services as companies like Block, Visa, Robinhood, and Mastercard use AI to reduce costs and increase operating efficiency. For fintech investors, the focus is moving beyond AI as a product feature toward AI as a driver of stronger margins and leaner operations.
Leopold Aschenbrenner's hedge fund, Situational Awareness, was reportedly forced to sell its public equity portfolio after leveraged bets on AI infrastructure stocks and shorts against software companies moved sharply against it. Citadel agreed to acquire the public positions as the fund's prime brokers worked to raise cash and meet margin requirements, marking a major test of Aschenbrenner's concentrated AI investment thesis.
Delays to US open banking rules are widening the payments innovation gap with Europe, where standardized APIs and instant account-to-account payments make it easier for fintechs to launch and scale. Without clearer mandates, US banks may remain dependent on costly integrations and insecure screen scraping, while global fintechs prioritize markets with better data access, lower processing costs, and more consistent regulation.
The next generation of banking will be built around AI agents that orchestrate end-to-end workflows and continuously make decisions, shifting banks from maintaining records to delivering measurable business outcomes. Rather than replacing core banking systems, institutions can layer agentic AI on top of existing infrastructure to automate lending, payments, fraud detection, and compliance while improving governance and reducing operating costs. The fintech opportunity is less about deploying standalone AI features and more about building the orchestration, data, and governance layers that enable banks to operationalize AI at enterprise scale.
Robinhood's Q2 earnings included data showing prediction markets overtook crypto and stock trading for the first time. Transaction revenue was $776m, up 44%. Options did $342m, event contracts did $156m, equities did $129m, and crypto did $100m, which fell 38% YoY. The company also posted record Q2 revenue of $1.31bn, up 32% YoY, with diluted EPS of $0.62 against a consensus nearer $0.42.
Rent payments fintech Flex has applied for an industrial loan company (ILC) charter, a move that would let it issue its own credit products instead of relying on sponsor banks. Flex lets renters split a monthly rent payment into smaller installments that better align with when they get paid, helping users avoid late fees without compounding interest or stacking loans. Since 2019, the company has processed more than $40 billion in rent payments for 3.2 million users and says it has helped customers avoid more than $780 million in late fees.
Fay is an AI agent that reviews fund administrator deliverables using more than 150 checks across financial statements, capital account statements, and supporting workbooks. The product combines AI-based document extraction with deterministic calculations to flag errors, reconcile figures across reports and periods, and reduce review time for CFOs and controllers while preserving source-level traceability and human judgment.
A recent NCAA rule change allowing corporate logos on college athletes' jerseys has created a new marketing opportunity for banks and fintechs. JPMorgan Chase's partnership with Ohio State and SoFi's deal with Notre Dame highlight how financial institutions are competing to reach students and young consumers through high-visibility sports sponsorships.
Ellis raised more than $10 million in seed funding led by First Round Capital to build an AI-native operating platform for private credit managers. The platform unifies data from fund administrators, loan-servicing systems, general ledgers, banks, legal documents, and spreadsheets, then uses specialized agents to automate reconciliation, LP reporting, portfolio monitoring, and compliance while keeping humans in control of approvals.
Increase founder and former Stripe engineer Darragh Buckley has acquired and rebranded a Washington community bank, giving his fintech direct control over regulated banking infrastructure instead of relying on partner banks. The move positions Increase to compete with banking-as-a-service providers like Lead Bank and Column, offering fintechs a more integrated platform that combines banking technology with a licensed bank while continuing to serve the local community.
Circle received a limited-purpose trust charter from the New York Department of Financial Services for Circle New York Trust, strengthening the regulatory foundation supporting USDC and its broader digital asset platform.
Most banks surveyed in a recent European payments study by BCG said they urgently needed to modernise their payments architecture, yet around one in five had already stopped or abandoned at least one modernisation programme.
QM is a multiplayer agent harness for companies. It remembers enough for people to work together without treating the whole company as one person with one permission set. The agent recognized personal, channel, group, team, and organization scopes, allowing people to collaborate through the same agent without becoming one undifferentiated user. QM is open-source and available for Slack and the web.
Model pricing can now go stale before a startup ships. OpenAI cut Luna's API price 80% only three weeks after launch and dropped Terra's price 20%. Anthropic released Opus 5 at half the price of Fable 5, while Google priced Gemini 3.6 Flash below Kimi K3 per task. Cheaper models are already forcing frontier labs to move before most enterprises have even started switching. If model spend matters to your margins, keep switching costs low.
2026.08.03·contraryresearch.substack.com·6 min read·TLDR Founders Service is what the customer experiences when trying to get something done. For people to trust you, you need to make their journey easy, respect their time, take ownership, ensure that advisable details inspire confidence, be competent and decent when something goes wrong, and build a system that learns. Trust takes time to earn but can be lost surprisingly quickly. It is one of the most valuable things a service provider can possess.
Forward-deployed engineers brought proximity, speed, and ownership of the outcome, but that isn't enough. Engineers are still often too far away from the key decision-makers who are critical for decisions and culture. Companies can't just rapidly hire and replace the entire C-suite in a month - they need outside help. Forward-deployed executives can help close the gap by bringing both technical and leadership experience to help businesses navigate the chaos and complexity of AI adoption.
QM is an agent designed for startups. Each employee gets their own isolated workspace and can work independently without affecting each other. They can also collaborate with the agent in channels, group messages, and projects. Skills are scope-owned and shareable by grant. QM can be used to search internal knowledge, build internal apps, work on existing repositories, and track projects in a shared channel with updates and follow-ups.
OpenAI's new campaign type, Agent, allows marketers to send ChatGPT searchers and users to a Business Agent conversation. The agents will be created through a combination of scraped data from the business website and instructions from advertisers. They will have access to product feeds and MCP tools, and can capture leads through custom forms. They can also launch ad campaigns that point to the agent rather than a website URL.
Most of what founders currently call a moat is a lease. Routing, distribution, marketplace liquidity, and the app-layer feature set all lose value as the models get better. What appreciates is proprietary data, expert decision-making, and a pipeline that turns those decisions into the next training run.
Startup options can turn into a bill the day an employee leaves. The usual 90-day window asks someone to pay the strike price and possibly taxes before they know whether the stock will ever be worth anything. That made more sense when startups exited in four to seven years, not today's 10 to 20 years. Backplanes gives people their tenure minus one year to exercise, while Craftwork gives them a full decade. If you use equity to recruit great people, make sure they can actually keep it.
Manifest ran an LLM router for four months across 7,000 users, then decided to shut it down.
WhatsApp is testing a feature that automatically moves messages from large businesses to an 'Offers & Updates' folder after a set number of hours.
Patrick Collison talks about dropping out of MIT twice, why founders should ask what happens if they succeed, and what Stripe's own data says about the best time to start a company.
Drive can come from pain, but it can also come from curiosity, high standards, or a family that never learned to coast.
Some quick tips on what investors look for before wiring money.
The FCC issued a ban this week on new, foreign-made humanoid robots, quadrupeds, robovacs, and sidewalk delivery robots, citing privacy risks from devices that map homes and public spaces plus supply-chain security concerns. China threatened retaliation, calling it a threat to "economic and trade stability." While the ban doesn't name China outright, the country controls roughly 85% of the global humanoid robot market. The rule doesn't target specific companies or actually curb data collection, and many US robotics firms that rely on Chinese components are worried it could gut their own supply chains instead.
Hyperscalers like AWS and Meta are snapping together factory-prefabricated modular datacenters, building mechanical and electrical fit-outs off-site like standardized Lego blocks to bypass construction labor shortages, compressing building schedules by 36% and cutting on-site licensed electrician labor by 85%. It's part of a broader trend of routing around old bottlenecks entirely: developers are increasingly wiring GPU clusters directly to co-located natural gas generation, small modular reactors, or existing nuclear plants, cutting time-to-power from 60 months down to under 18.
A developer got a 28.9M parameter language model running entirely on an ESP32-S3, a microcontroller with just 512KB of SRAM. The trick, borrowed from Google's Gemma models, is called Per-Layer Embeddings: instead of loading the full network into fast memory, about 25 million parameters live in flash as a lookup table, and only the bytes needed for the current token get read out during inference, while the smaller layers stay in SRAM. After quantization, the whole model shrinks to 14.9MB, fitting comfortably on the board's flash, and it generates about 9.5 tokens per second on an $8 chip.
Qualcomm finished buying Modular, the AI infrastructure company behind the Mojo programming language and the MAX inference framework. Modular's whole appeal is that its software lets the same model code run across CPUs, GPUs, NPUs, and custom silicon without rewriting kernels for each one. Qualcomm just raised its 2029 target to $40 billion in non-handset revenue, with data center alone expected to contribute over $15 billion, and it's betting that owning the compiler and serving layer outright, not just the silicon, is what gets new Qualcomm AI chips performing well from day one.
Sponsored
General-purpose AI and data tools aren't built for physical systems. The result? Frequent breakage under pressure, at speed, and on high-consequence projects. Nominal [8] connects the major product components engineers now stitch together to help hardware engineers validate, operate, and improve machines. See how Anduril engineers
Robots trained on step-by-step instructions often fail on new combinations of familiar steps they weren't directly shown, and collecting demonstrations for every possible combination is impractically expensive. This paper breaks that failure down into three distinct causes - shifts in individual instructions, shifts in how instructions combine, and shifts in the surrounding context - to figure out exactly what a training set needs to cover rather than how much of it. Full coverage of every instruction combination isn't necessary: a carefully chosen quarter of the task space can match it, as long as that subset preserves the real dependencies between actions rather than just varying surface details.
Tau Robotics has opened an invite-only San Francisco pilot using humanoid robots for vacuuming, surface cleaning, trash removal, and deep cleaning around appliances. The robots combine AI with live human supervision, with each unit assigned to one appointment at a time during the initial roll-out. The service is priced at $30 per hour and is currently limited to selected households.
A review of over 80 Chinese academic papers and patents found military and security-linked researchers using "model distillation" to train smaller domestic AI systems on outputs from OpenAI and Anthropic models, letting them extract frontier-level capability without needing the massive GPU clusters normally required to train from scratch. It's a workaround that directly undercuts the logic of US chip export controls, since restricting access to advanced silicon matters less if capability can be distilled down onto whatever hardware is already available domestically.
GPU spot rental rates could surge 10x–15x as frontier models gain human-level software engineering capabilities. Physical GPU supply only scales ~3x annually due to fab and packaging bottlenecks, so the immense economic value of automated labor will drive inference willingness-to-pay far higher than current market spot prices.
Gritt raised a $26 million Series A, bringing total funding to $34 million, to build AI-controlled robots using off-the-shelf hardware that install solar panels roughly 4x faster than a human crew.