GPU compute is expensive and orchestrating training jobs across cloud providers is a pain. TensorPool is a CLI that handles the infrastructure so ML engineers can stop being part-time DevOps.
Over 300,000 apps built on the platform, 200,000 users, and a pricing page that goes up to $800 a month. a0.dev is turning React Native app development into something closer to a conversation than a codebase. The question is whether AI-generated apps can hold up when real users start filing bug reports.
Software engineers got GitHub, VS Code, and Cursor. Hardware engineers got... spreadsheets and PDFs. Artifact is building a collaborative IDE for electrical systems design, and they've already got Boom Supersonic and K2 Space as customers. This team built avionics for hypersonic aircraft. They know exactly what's broken.
Design tools and code editors have been separate universes for decades. Onlook merges them into one visual editor that writes directly to your React codebase. With 22,000 GitHub stars and growing, this might be the tool that finally kills the design-to-dev handoff.
Running transformer models on edge devices usually means accepting terrible performance or buying expensive hardware. Exla's SDK aggressively quantizes models for up to 80% memory reduction and 3-20x faster inference with a few lines of code.
Every developer has wasted a week integrating an API that should have taken a day. ReJot wants to replace the SDK-plus-backend-wiring pattern with reusable building blocks that include the data layer. Two ex-Adyen engineers from Amsterdam who spent years thinking about distributed data.
15,000 users, a million modified pages, and a premise that basically says: you know better than the product team. Tweeks.io turns your complaints about websites into actual browser modifications, no code required.
Systems engineering tools haven't kept up with the systems they're supposed to engineer. Dalus wants to replace the spreadsheets and siloed documents with something that actually works in real time.
Every company has a graveyard of SQL scripts, cron jobs, and one-off connectors that nobody wants to touch but everything depends on. Superglue replaces them with AI-powered, self-healing integrations.
Documentation drift is the silent killer of developer experience. Promptless monitors your PRs, Slack threads, and support tickets, then updates your docs automatically. No prompting required.
AI agents are stuck inside chat windows. A1Base wants to give them phone numbers, email addresses, and WhatsApp accounts so they can operate in the real world.
Most AI tools target knowledge workers who write emails. Macadamia is going after mechanical engineers, construction firms, and manufacturers with AI agents that actually understand CAD files.
Live voice translation has been a sci-fi promise for decades. Pinch is building the API and desktop tools to make it work in actual meetings, with actual tone preservation.
Enterprise search has been mediocre for decades. ZeroEntropy is building an AI-native search API that handles the messy, unstructured documents most RAG systems choke on.
Every AI product needs a chat interface. Building one from scratch takes weeks and the result is usually mediocre. assistant-ui is an open-source component library that solves this in hours.
The old coding interview tests whether you can solve algorithms on a whiteboard. Outship tests something different: can you actually ship software with AI tools? That might be a better signal for how engineering actually works now.
Electronic component discovery is stuck in the 1990s. Four ETH Zurich grads are building natural language search across millions of parts, and they have bigger plans from there.
Most AI agent tasks are repetitive. Butter records tool call trajectories and replays them deterministically, cutting out the LLM entirely for cases that could have been a script. Faster, cheaper, and more predictable.
Four college friends from Cornell and Stanford are building an AI-powered game engine. The ambition is big. The question is whether game developers actually want AI touching their creative process.
Documentation is where good engineering teams go to feel guilty. Quantstruct automates the whole thing by watching your codebase and writing the docs for you.
AI agents need access to your tools to be useful. Corsair is the permission layer that makes that access safe enough to actually turn on.
Most EV startups want to build the car. Steinmetz wants to build the thing that makes the motor spin, and they claim 3x the power density of existing solutions.
AI agents are shipping to production faster than teams can debug them. Subtrace traces every HTTP request and tool call without touching your code, giving you a real-time window into what your agents are doing and why they fail.
Every developer building AI agents is solving the same infrastructure problems from scratch. Truffle wants to hand them a solved stack so they can focus on the actual agent logic.
Structural and thermal analysis in aerospace and defense takes weeks and requires specialists. Arda built an AI that delivers first-pass calculations in minutes, letting engineering teams run analysis 10x faster.
Most AI support bots hallucinate answers your docs never contained. Inkeep grounds everything in your actual content and then lets you build agents on top of it. The customer list suggests it works.
Kafka is powerful, popular, and notoriously difficult to work with. Quix built a platform that lets engineering and data teams develop, deploy, and monitor stream processing pipelines without wanting to throw their laptops out the window.
The terminal is having a moment, and Cline CLI 2.0 is betting it can be the thing that makes AI coding agents actually useful in production — not just in the IDE.
Ask Ellie wants to be the engineering team's single source of truth — and it's betting that source of truth should live in a chat box.
Every AI agent needs to talk to the outside world. Klavis AI is building the managed infrastructure layer so developers don't have to wire up OAuth flows and tool integrations from scratch.
AI coding agents are getting good at writing code — turns out the part that keeps breaking is everything underneath it.
ZenMux wants to be the last LLM integration you ever write — and it's willing to pay you if it isn't.
Vibe coding is great until your AI assistant stares at a stack trace like a golden retriever watching TV.
SQLite for the web is a solved problem with about five solutions. Bunny Database is betting its CDN roots make it the one that actually sticks.
Maige automates the grunt work of GitHub issue management with AI that labels, assigns, reviews, and responds. Built by Rubric Labs with 4,300+ repos already using it.
Codex-Spark hits 1,000+ tokens per second by running on Cerebras chips — and that partnership might matter more than the model itself.
An AI builder that also launches your product to a discovery feed and optionally staples a Solana token to it is either the cleverest product thesis of 2025 or a three-way stretch that snaps under pressure.
Your AI coding sessions vanish into terminal logs nobody can read — one open-source tool thinks that's worth solving.
Telnyx just shipped a voice layer for Clawdbot — and the actual question is whether talking to your agent beats typing at it.
ClawMetry is a free observability dashboard for OpenClaw agents — and the fact that it needs to exist says a lot about where AI tooling actually is right now.
An open-source model from a Chinese AI lab just posted benchmark numbers that should make Anthropic at least glance up from whatever they're doing.
Everyone's shipping AI code review tools — Unblocked's bet is that context is the whole game.
Turning Google Sheets into a REST API isn't a new idea, but Sheetful is betting there's still a real market for doing it cleanly and for free.
Zero config, zero login, zero patience required — PinMe is betting that the hardest part of shipping a frontend is everything that comes before the actual shipping.
Building an AI agent is the easy part now — deploying it somewhere useful, connecting it to your users, and keeping it running is where things quietly fall apart.
Knowing what software a company runs is only valuable if the data is fresh, sourced, and queryable — which is exactly what most technographics providers have historically failed to deliver.
Postgres scaling has always meant painful tradeoffs. PgDog, built by the engineer who sharded Postgres at Instacart during peak pandemic growth, wants to make horizontal scaling something you drop in rather than something you dread.
When your codebase is millions of lines spread across hundreds of repos, your IDE's search bar is useless. Sourcegraph built the tool that makes massive codebases navigable, and now they're layering AI on top of it.
Embedded finance had its moment with payments and lending. Insurance is next, and Maple is building the API layer that lets any company offer insurance products without becoming an insurance company.
When every AI coding tool is racing to be smarter, one open-source project is betting the real problem is that nobody taught these agents how to think in steps.
Your data pipelines are broken more often than you think. Acceldata built an observability platform that monitors data quality, pipeline health, and infrastructure costs so you find out before your CFO does.
Every AI analytics tool promises to answer your data questions. Relta says the others are guessing and it can prove its answers are correct through formal verification of generated SQL.
Your AI app can talk to one data source at a time. Airweave wants it to search all of them at once, and the pitch is more practical than it sounds.
The team that built Gatsby.js is back with an open-source TypeScript framework for AI agents. 22,000 GitHub stars suggest developers are paying attention.
The idea that a single script tag could turn a dead SaaS landing page into a living, breathing room full of avatars is either the best idea in engagement marketing this year or a very confident bet on collective nostalgia.
Engineering teams spend 20 hours waiting for code reviews. Greptile indexes the whole codebase and cuts that to under 2. Over 1,000 teams already use it.
OpenClaw is everywhere. Running it without wanting to throw your laptop out a window is a different problem entirely.
Running one AI coding agent is a productivity trick. Running five simultaneously without losing your mind is a different problem entirely, and Superset is betting it's the one worth solving.
The pitch is almost too simple: describe what you want from any website, and Anything API ships you a working endpoint.
The supply chain attack problem is real, unglamorous, and quietly getting worse — and Koidex just built a search bar for it.
The pitch is almost annoyingly simple: a fully managed AI agent host for under four bucks a month, because someone finally did the math on server margins.
Every AI product right now is a chatbot that spits out walls of text, and OpenUI thinks that's a solvable problem.
Everyone's chasing the frontier. Google quietly built something for the people who have to pay the bill.
The dirty secret of most teams using OpenClaw right now is that it's one person's setup, running on one person's machine, and everyone else is either SSH-ing in awkwardly or just not using it.
The guy who made developers care about security is back, and this time he's betting that AI agents are only as good as the skills you feed them.
Managed infrastructure for AI agents is quietly becoming its own product category, and Donely is betting you'd rather pay nothing upfront than touch a terminal.
Code review tooling has been a solved problem on paper and a daily annoyance in practice for years, and a small Mac app wants to fix that without asking you to change how you work.
The best version of Hacker News has always been the one someone else builds.
Everyone's building smarter agents, but nobody wants to admit that 'smart' doesn't mean much if the agent forgets everything the moment you close your terminal.
A social network where the users are AI agents, the posts are written by bots, and your job as a human is to sit back and watch.
VibePad maps your PlayStation controller to Claude Code and Codex shortcuts, and I'm annoyed at how much sense it makes.
Running GPU servers for AI models is expensive when they're busy and wasteful when they're idle. Cerebrium thinks serverless is the answer, and the economics actually make sense.
Building on Solana has always meant fighting your local environment before you write a single line of real code, and Surfpool thinks it has a cleaner fix than anything currently on the table.
Every B2B SaaS app eventually needs comments, presence indicators, and notifications. Velt thinks building those from scratch is a waste of your engineering team's time.
Vibe coding meets mobile development. Bloom gives you a backend, cross-platform builds, and native testing on your phone. The catch is that 'no-code' still has to produce code that works.
AI coding tools have gotten impressively good at writing code and impressively bad at producing output you can actually paste anywhere.
Background jobs are the plumbing of every serious application. Trigger.dev thinks the plumbing should be TypeScript-native, open source, and way less annoying to debug.
Clidey started with WhoDB, an open-source platform for modeling organizations. Now they have added Docucod, which generates and maintains documentation straight from your codebase. Two products, one team, and a bet that data-driven decision-making needs better infrastructure.
The pitch is simple: what if your community platform wasn't someone else's product?
Formal verification has always been powerful and almost completely impractical. Theorem is training models to make it 10,000 times faster, and the early results on GPU code and cryptography are hard to ignore.
Bad documentation doesn't announce itself. It just quietly breaks things until someone has to read it.
The pitch is seductive: just say what you want and the AI figures out the rest. Whether that holds up past the demo is a different question.
Everyone I know who's tried to grind DSA has the same story: two weeks of enthusiasm, one hard graph problem, and then the tab quietly closes forever.
AI agents are shipping faster than anyone can test them. Janus is building simulation environments that catch the failures before they reach production. The pitch: trust, but verify.
Two brothers, one from SpaceX and one from a clean energy startup, are using AI to match complex weld jobs to the right fabrication shops. The pitch: cut bid and scheduling time by 80%.
Three repeat founders who scaled a company to $400M valuation got tired of status meetings. So they built a tool that reads every line of code and tells you what actually shipped. Six-figure ARR in a week.
Morphik is an open-source retrieval platform that gives AI apps access to real documents instead of confident fabrications. Space-tech teams and brokerage developers are already using it.
Describe the prediction you want to make, connect your data, and Plexe builds, evaluates, and deploys the model for you. That is either the future of ML or a very good demo.
Auth0, Clerk, and Supabase Auth all want to host your authentication. Better Auth thinks you should own it yourself. With 13K GitHub stars and 350K monthly npm downloads, developers seem to agree.
Computer-use agents need somewhere to work that is not your laptop. Cua provides sandboxed cloud desktops where AI agents can see screens, click buttons, and run applications without touching your actual machine.
End-to-end testing is the part of software development that everyone agrees is important and nobody wants to do. Docket replaces brittle CSS selectors with coordinate-based AI agents that see your app the way a human does, and the tests actually stay green.
Researchers spend more time cleaning data than analyzing it. Percival writes the code, suggests the analysis, and interprets the results, all inside your existing workspace. It is what Jupyter notebooks would be if they were invented today.
AI code generation is shipping features faster than ever. It is also shipping bugs faster than ever. Synthetic Society deploys swarms of AI users that behave like real humans to find the UX flaws, broken flows, and edge cases that manual QA misses. Two MIT and Columbia dropouts are betting that testing needs to evolve as fast as the code it checks.
Everyone is building AI agents. Almost nobody knows how to make them reliably better. ZeroEval closes the feedback loop between production failures and prompt optimization with calibrated LLM judges that learn from your corrections, turning weeks of manual tuning into hours of automated improvement.
Every digital product generates conversation. Support tickets, in-app chat, feedback forms, onboarding flows. Costream is building infrastructure to capture, index, and make all of that dialogue searchable and actionable. Two Cornell dropouts who already scaled an AI product to 100K users think they can turn product conversations into structured intelligence.
GitHub has become the default. That is the problem. gitdot is not building a GitHub clone with more features. They are building an opinionated platform for open-source maintainers who care about code quality over vanity metrics. No AI copilot. No free private repos. No stars. A Rust-powered Git server and a CI/CD platform designed for people who treat code as craft.
Half of all search volume could shift to AI interfaces by 2028. That is a $1.3 trillion advertising opportunity with no incumbent ad network serving it. Imprezia, built by a former Amazon Ads and Meta Ads engineer, is placing contextual brand integrations inside AI chatbot conversations. One-line SDK. No banner ads. Just relevant brand mentions woven into natural dialogue.
Every AI team knows they need better evals. Most of them are still building annotation pipelines by hand, burning weeks on infrastructure that has nothing to do with their actual product. MangoDesk generates custom eval and post-training data pipelines in seconds. That is either a huge time-saver or a huge shortcut, and the difference matters.
Manual QA is slow. Selenium scripts break every sprint. Autosana lets you write end-to-end tests in plain English for iOS, Android, and web apps, then runs them on cloud-hosted agents that heal themselves when the UI changes. Two former mobile startup engineers think the entire testing layer is about to get replaced.
Three former Apple engineers who co-created SwiftUI built a tool that lets you describe an iPhone app in plain English and get production-ready Swift code you can ship to the App Store. No Xcode required. No Mac required. You build the app on your iPhone.
Every productivity app you use has its own silo of context about you, and none of them talk to each other. Epicenter stores everything in plain text and SQLite files you own, then lets local-first apps share that memory. It is the anti-cloud-lock-in bet, and the founder has 10,000 commits a year to back it up.
Browser automation breaks constantly because websites change. Notte built a full-stack browser infrastructure platform that lets AI agents run on the internet at production scale, with under 50ms latency, 1000+ concurrent sessions, and an agent success rate above 90 percent. Two MIT researchers who spent years studying why web automation fails think they finally fixed it.
Production debugging is still mostly manual. An engineer gets paged, opens Datadog, stares at logs, traces the issue across three services, and writes a fix four hours later. TraceRoot built an open-source AI agent that connects to your telemetry, traces the root cause, and drafts the pull request. The SDK has over 10,000 downloads and the founding team fixed 300+ production bugs at Meta and AWS before deciding to automate themselves out of a job.
Every AI coding tool is built for web developers. Firmware engineers, the people who write the software that runs inside cars, medical devices, and industrial equipment, have been completely ignored. Embedder is an AI-powered IDE that reads datasheets, generates production-ready drivers, and tests code on real hardware. Tesla, Samsung, and Medtronic are already using it.
The Model Context Protocol is becoming the standard way AI agents connect to the outside world. Manufact, formerly mcp-use, built the open source SDK that 4,000 companies already depend on. Now they are building the cloud infrastructure layer on top. With NASA, NVIDIA, and SAP as customers and 5 million downloads, this three-person team from Zurich and San Francisco is positioning itself as the default MCP platform.
Jira has been the default ticket management system for engineering teams for fifteen years, and almost nobody likes it. Janet AI is not adding AI features to Jira. They built an entirely new system where tickets create themselves from Slack conversations, meeting transcripts, and emails, then update themselves when PRs merge. Two Cornell and UW-Madison grads who were founding engineers at previous YC startups think the entire concept of manual ticket management is about to die.
Whisper works great for English. For Urdu, Bengali, and Greek, not so much. Uplift AI is building foundational voice models for regional languages and delivering them through APIs and SDKs so developers in emerging markets can build voice-first applications that actually work.
Billing logic for AI products is a special kind of painful. Usage-based pricing, credit systems, feature gating, plan migrations. Autumn puts all of it in a layer over Stripe so you stop building billing code from scratch.
Every major AI model today requires a data center to run. DeepGrove is building technology to bring frontier-level intelligence to edge devices, from phones to embedded systems, without sacrificing capability. If they pull it off, the implications for privacy, latency, and accessibility are enormous.
Most coding agents lose the plot halfway through a long task. Magnitude splits the work across specialized subagents for planning, reviewing, debugging, and browsing so the main agent stays focused on what you actually asked for.
AI coding agents are good at writing code but terrible at knowing what libraries actually do, what APIs have changed, and what documentation says right now. Nozomio indexes external context so your agents stop hallucinating about frameworks they last saw in training data.
Legacy Windows applications run the backend of healthcare, finance, and logistics. Nobody builds integrations for them because nobody wants to. Cyberdesk trained an AI agent that learns how to use these apps the same way a human would, and hospitals are already using it for patient intake.
End-to-end tests are brittle, slow to write, and nobody wants to maintain them. Lark replaces Cypress and Playwright scripts with plain English descriptions and uses AI to figure out the selectors, waits, and assertions.
AI coding agents are powerful but chained to your laptop. Omnara cuts the cord by letting you monitor, review, and control Claude Code and Codex from your phone or any browser.
Every AI coding tool targets software developers. SigmanticAI built one for chip designers, automating the verification workflow that eats 70% of the semiconductor development cycle. Siemens is already a partner, and the fine-tuned Verilog models are beating Cursor on accuracy benchmarks.
Everyone wants to build AI agents that use computers like humans. The problem is you cannot train those agents on production systems without breaking things. Halluminate builds realistic sandbox environments that replicate Salesforce, Slack, and enterprise tools so AI labs can train and benchmark computer use agents safely.
Everyone is fighting over which model to run. Luminal is fighting over how fast you can run any of them. Their ahead-of-time compiler turns AI models into optimized GPU code and is already beating vLLM and TensorRT-LLM on throughput benchmarks. Three people, $5.3 million, and a very different theory of how inference should work.
AI agents are shipping into production with no way to safely test them against real-world failures. Playgent builds high-fidelity sandbox environments where teams can reproduce bugs, run evaluations, and train agents through reinforcement learning, all without touching live systems.
A bad dimension on an engineering drawing can cost a manufacturer millions in scrap and delays. F4 automates GD&T compliance checks using AI, built by engineers who have lived through the pain at Tesla and SpaceX.
No-code tools have been promising non-technical founders they can build real software for years. Floot thinks the problem was never the interface. It was the framework underneath.
AI coding tools are powerful but disconnected from what you actually see on screen. Stagewise puts a coding agent directly in your browser so you can click on elements and tell it what to change.
Every AI shopping agent needs a product database. Channel3 built one that covers millions of products from thousands of brands, with sub-2-second lookups and built-in monetization.
Coding agents are getting good at writing code. They are terrible at deploying it safely. Proliferate gives them sandboxed environments, event triggers, and verification workflows so they can ship without human babysitting.
Topology optimization is one of the most compute-intensive tasks in mechanical engineering. Topological built a foundation model called UToP that does it nearly 2,000 times faster than conventional methods with less than 5% error.
The prompt-to-app space has a dirty secret: most tools generate frontends that look great and backends that are held together with duct tape. VibeFlow is trying to fix that by making backend logic visual, editable, and actually understandable.
AI agents fail in ways that are hard to reproduce and harder to diagnose. Fulcrum builds red-teaming agents that find the failures for you, uncover environment bugs, and catch reward hacking before it ships.
Cursor and Copilot built AI coding for software engineers. Lotas built it for data scientists who live in RStudio, write R, and have been completely ignored by the coding assistant wave.
Forward deployed engineers are the most expensive people at AI companies, and most of them track their work across Slack threads, Linear tickets, and memory. Nixo centralizes the chaos.
Everyone is building AI agents. Almost nobody can test them properly before deployment. AgentHub provides the simulation environments, tracing tools, and evaluation frameworks that agent developers need.
Security awareness training has been stuck on fake phishing emails for a decade. GhostEye builds exposure maps of employees and launches autonomous simulated attacks across email, voice, SMS, and deepfakes.
Firmware engineering is one of the last software disciplines untouched by AI. BootLoop changes that with an agent that reads datasheets, writes production-ready embedded code, and validates it against actual hardware using oscilloscopes and logic analyzers.
AI coding agents have full access to your file system, your secrets, and your network. ContextFort gives security teams a way to see exactly what those agents are doing, without trusting the agents to self-report.
Nobody reads the documentation. Nobody watches the full training video. Golpo converts your PDFs and docs into short whiteboard explainer videos with AI narration, betting that animation and a human voice can make boring content actually consumable.
Every developer has built a weekend project that works locally but would take three weeks to make production-ready. Modelence ships everything you need to skip the boilerplate and go live from day one.
Monitoring tools tell you something broke. Interfere tells you what broke, why it broke, and how to fix it before anyone files a ticket. That is a meaningfully different product.
Every sales engineer has experienced the empty-state demo disaster. Eden generates production-quality synthetic data on demand so your product looks like it has actual customers in it, without touching production databases or spending a week building seed scripts.
Every AI company has a GPU performance problem. Most solve it by hiring expensive kernel engineers. Wafer automates the entire workflow, from profiling to patching, and claims 1.5 to 5x speedups without anyone touching CUDA.
Building an AI agent is getting easier. Deploying one into production, with persistence, scaling, safety guardrails, and monitoring, is still a multi-week engineering project. Arcten wants to compress that into minutes.
The no-code web app space is crowded. Fastshot thinks the real gap is mobile. Describe your app in plain English, get a production-ready React Native build for iOS and Android. The question is whether that gap stays open long enough.
Every AI assistant lives in a browser tab. Logical thinks that is the wrong model. Their desktop copilot watches what you are doing across apps and proactively helps without being asked. It is the Clippy pitch done right: context-aware, privacy-first, and actually useful. The question is whether people want an AI that is always watching.
GPU costs are the single biggest line item for most AI labs. SF Tensor claims it can cut those costs by up to 80% by automatically sourcing cheap GPUs across providers and optimizing kernels for any hardware. Three brothers, all CS graduates before turning 20, are betting that vendor lock-in is the real bottleneck in AI research.
Coding agents can write code. They cannot deploy it. Specific gives agents a cloud platform where they can spin up databases, configure secrets, and ship to production without a human touching the infrastructure.
Voice assistants have been promising to replace your keyboard for a decade. Caddy is the first one built by people who actually understand productivity software, and it shows.
Vibe coding means waiting for AI to generate your code. Clad Labs figured out that developers spend that dead time on social media anyway, so they built social media into the IDE. It is either genius or an abomination.
Seventy percent of Fortune 500 companies still run mission-critical systems on COBOL. The engineers who built them are retiring. Hypercubic is racing to capture their knowledge before it walks out the door.
Every AI agent forgets everything between sessions. Hyperspell builds the memory layer that connects agents to Slack, Gmail, Notion, and Drive so they actually know who you are and what you are working on.
Enterprise codebases are too big for anyone to understand. Sourcebot gives both humans and AI agents the ability to search, query, and reason across millions of lines of code without sending anything off-premises.
Voice agents and chatbots need fast retrieval. Like, really fast. Moss built a Rust and WebAssembly search runtime that runs locally on any device and returns results in under 10 milliseconds. No database server, no network hop, no excuses. Grammarly and HubSpot are already customers.
Everyone is racing to generate code faster. Scott AI is the rare product that wants you to slow down first. Spec before codegen. Debate before commit. It sounds counterintuitive until you realize how many production fires start with 'nobody reviewed the architecture.'
If you have ever tried to run three coding agents in parallel on the same machine, you have hit the port clash wall. Coasts fixes that with localhost isolation per git worktree, and it is free.
Kubernetes incident response is a nightmare of runbooks, alert fatigue, and 3 AM pages. Kestrel AI deploys agents that find root causes and generate production-ready fixes automatically.
Kafka is overkill for most streaming use cases and everyone knows it. s2.dev treats streams as a storage primitive with a REST API, sub-50ms latency, and no cluster to manage.
Cleft embeds feature flags directly in your codebase. No third-party SDK, no external dashboard, no vendor lock-in. Just flags that live where your code lives.
LiveBench releases new benchmark questions every month so AI models cannot memorize the answers. Eighteen tasks across six categories, all with verifiable ground-truth answers. No LLM judges. No stale data.
Tusk is an AI agent that reads your Jira tickets, understands your codebase, writes tests, fixes UI bugs, and opens PRs. It does the work that every engineering team knows is important but nobody wants to do.
AI agents are everywhere and nobody can tell you why they broke. Fernstone gives developers tracing, evals, and cost tracking for autonomous AI systems so you can actually debug the thing before your customer does.
Your CI pipeline is rebuilding code that has not changed, and you are paying for every wasted minute. Trace Machina's NativeLink is an open-source remote execution and caching system that handles over a billion requests a month for companies like Samsung.
Most AI agent platforms give you a chat window. Helix gives each agent a full GPU-accelerated 4K desktop with a browser, terminal, and filesystem. It is a fundamentally different architecture and I think it is the right one.
While React frameworks grab headlines and funding rounds, Quasar has been building a cross-platform development toolkit on Vue.js that ships web apps, mobile apps, desktop apps, and browser extensions from a single codebase. And it actually works.
I have been waiting for someone to build the air traffic control tower for coding agents. Emdash is that tower, and it is open source.
Four Swedish founders looked at the patent industry and saw attorneys manually reading thousands of documents. They built an AI system that does it faster and shows its work.
Tapeout is the most expensive deadline in technology. Visibl Semiconductors is building AI agents that keep chip design teams from missing it.
Deploying a coding agent to run overnight sounds great until you realize there is nowhere good to put it. Terminal Use wants to fix that.
On-call engineering in financial services is brutal. Corelayer puts an AI agent on it that monitors, debugs, and cites its sources. The $2.8 million annual savings claim got my attention.
Most AI teams cannot explain why their models behave the way they do. Envariant is building the SDK that lets you look inside, understand, and steer model behavior.
Moda is the monitoring layer for AI agents, surfacing patterns across hallucinations, tool call failures, and behavioral drift that traditional logging completely misses.
Sentrial is the monitoring platform for AI products in production, detecting real-time when agents hallucinate, loop, or frustrate users, then diagnosing root causes and recommending fixes.
Synthetic Sciences is building the infrastructure for AI co-scientists, starting with an agent platform that delegates literature reviews, trains models on cloud GPUs, and generates publication-ready LaTeX.
Compresr provides an API that compresses LLM context up to 200x without losing what matters, cutting token costs and improving accuracy for agents and RAG systems.
Lucent is an AI product manager that watches user sessions, detects bugs and UX issues, and automatically suggests improvements based on how people actually use your product.
Chamber orchestrates GPU infrastructure with AI agents that detect unhealthy nodes, forecast demand, and reallocate resources so ML teams can run 50% more workloads on the same hardware.
Crow plugs an AI agent into existing applications so users can accomplish tasks through natural language chat instead of navigating complex menu structures and workflows.
Skillsync creates structured skill profiles from developers' open source contributions on GitHub, letting companies search for specific capabilities like 'deployed custom neural nets on edge' and find engineers already building in their domain.
The Token Company provides a drop-in API that compresses LLM inputs by 66%, cutting costs by 3x and improving response times by up to 37% while maintaining or improving output quality.
Velum Labs monitors production query traffic to detect when teams calculate the same metric differently, traces root causes through query lineage, and generates enforceable data contracts to prevent future breaks.
Zatanna observes a human performing a software workflow once, reconstructs the underlying request behavior, and exposes it as a production-grade API endpoint for AI agents and internal systems.
Syntropy is an autonomous coding agent that takes a feature description and produces a fully tested, production-ready pull request while you do other things, sending updates via Slack along the way.
Glue is an open source design canvas that integrates with coding agents like Claude Code and OpenAI Codex, letting developers leave visual feedback on their app and watch agents convert it to code in real time.
Aurorin built a new parametric CAD kernel from scratch to be fast, stable, and AI-native, so mechanical engineers can design complex parts using both traditional tools and natural language in a single interface.
Approxima automates end-to-end testing by opening your application in a real browser on every pull request, checking that changes actually work without requiring test scripts or maintenance.
IncidentFox autonomously investigates production incidents by querying logs, metrics, and deployment history, delivering root cause analysis and one-click remediation scripts that wait for human approval.
Sonarly connects to your monitoring stack and turns noisy alerts into root-cause analysis and ready-to-merge pull requests. It is trying to be the on-call engineer that never sleeps.
Polymath builds complex multi-tool environments for training AI agents through reinforcement learning. Their Horizon-SWE benchmark tests what happens when you ask AI to do real work across real tools over long time horizons.
AI agents like OpenClaw can do almost anything, including leak your SSN, execute malicious code, and fall for prompt injections. Clam puts a semantic firewall around them to stop the bad stuff.
AI can see and understand the physical world, but building real-time vision applications is painful. Overshoot provides the infrastructure so developers can build security, safety, robotics, and consumer vision products without reinventing the stack.
Ashr generates synthetic user journeys across voice, text, images, and tool calls to test AI agents before they hit production. Prompt versioning with inline diffs shows you exactly what changed and why it broke.
21st.dev started as the largest React component registry for AI apps with 1.4M developers. Now they are building a full SDK for deploying AI agents with built-in UI, memory, observability, and sandboxed execution.
RunAnywhere provides an SDK for deploying AI models on iOS, Android, and edge devices. Their MetalRT engine runs 1.67x faster than llama.cpp on Apple Silicon. 10K+ GitHub stars and growing.
APIs were not built for agents. Orthogonal gives developers and AI agents instant access to hundreds of APIs through MCP or SDK with pay-as-you-go billing. No API key management, no billing headaches.
Building a short-form video feed used to require a team of 20 engineers and a year of development. ShortKit gives you the same capability as a drop-in SDK with ML-driven pre-fetching, enterprise-grade transcoding, and native ad integrations.
Mendral is an AI DevOps engineer built by the people who wrote Docker's first lines of code, and it automatically diagnoses CI failures, catches flaky tests, and opens pull requests with fixes before your team even sees the red build.
Salus is an API that wraps around AI agents to inspect and block incorrect actions before they execute, preventing the kind of costly mistakes that evals and observability tools only catch after the damage is done.
Superset is a free, open-source terminal app that orchestrates multiple AI coding agents in parallel, each running in isolated Git worktrees, so developers can delegate ten tasks at once without agents stepping on each other.
Arga Labs provides sandbox infrastructure with fully functional mocks of external services so you can validate code changes and AI agent behavior at scale without risking production data or hitting real APIs.
Manicule is a documentation agency that uses AI agents for code verification and auditing at machine scale while human writers handle strategy and quality, already at $15K MRR with clients like Greptile, Reducto, and Rootly.
Primitive is building the communication infrastructure that fully autonomous agents need to interact with the outside world, handling the email, messaging, and notification layers that agents cannot currently access on their own.
The most annoying part of using an AI coding agent is babysitting it. Anthropic just decided to fix that, and the fix is either very smart or a little terrifying depending on how much you trust classifiers.
Training data has always been the unglamorous part of AI, and somebody finally built a Python SDK around fixing that.
Managing MCP servers across a dozen AI clients is exactly as painful as it sounds, and someone finally built the obvious fix.
Replicas lets engineering teams delegate tasks to AI coding agents that run in sandboxed VMs with full development environments, accessible from GitHub, Slack, Linear, or a web dashboard, and reportedly shipping over 30% of pull requests for some teams.
AI voice agents are shipping into production call centers and nobody has a clean way to know if they're actually working.
A Chinese AI lab just launched a model that reportedly participated in its own evolution, and the demo is convincing enough that I can't just scroll past it.
The idea of running a full emulation of your cloud setup before a single dollar gets spent in prod is either the most obvious thing in the world or something the industry somehow forgot to build.
I've watched developer friends rack up hundreds of dollars in Claude API costs across four different tools without realizing it, and nobody has really solved that until now.
Every Claude Code power user has hit the wall. Edgee thinks a token compressor running at the edge is the fix, and their benchmark numbers are harder to dismiss than I expected.
Remodex lets you run Codex from your couch, your commute, or wherever you left your laptop charging in the other room.
Everyone is building voice agents. Almost none of them sound like anything a human would want to talk to.
Someone built a leaderboard for AI token consumption and I am genuinely unsure whether to admire it or be alarmed by it.
Vision-enabled AI agents that browse the web autonomously is a real pitch in 2025, and somehow the vibe is ASCII art and cat puns.
The most annoying part of AI-assisted UI work isn't the code, it's explaining what you're looking at.
