Harper

We’re hiring a Forward Deployed Engineer. Harper works with some of the world’s largest enterprises to build and deploy production systems at meaningful scale. We’re looking for an experienced technical leader who can work directly with customers and partners to turn complex requirements into working solutions. You’ll move between architecture, development, technical delivery, presales and partner enablement—helping customers understand what is possible, then building it alongside them. The role also offers the opportunity to spend your days working with Harper, a unified platform that brings application logic, data, caching and messaging together in one distributed runtime. You’ll help organizations improve performance, reduce operational complexity and deliver technology at scale. The ideal candidate brings deep enterprise experience, a strong development foundation, AI fluency and the ability to own customer problems from the initial conversation through successful deployment. This is a remote, high-ownership role for someone who wants to build consequential systems and help shape an emerging category of software infrastructure. Learn more and apply: https://lnkd.in/g-fdiPRZ #ForwardDeployedEngineer #EnterpriseSoftware #NowHiring #Hiring

Web personalization is a platform decision. The architecture behind it determines whether teams can use live data, where application logic runs, how many systems must stay synchronized, what global latency looks like, and how costs behave as traffic grows. After benchmarking Harper against a Vercel-based architecture on personalized data, we wanted to examine the broader landscape: What are the most common ways organizations architect personalization, and what does each approach optimize for? We evaluated five architectures: → Static variant generation → Edge personalization → Server-side rendering with a separate data tier → Client-side personalization → A unified runtime We then compared them across the factors that matter at the systems level: → Data freshness → Latency → Personalization depth → Operational complexity → Cost at scale → Crawler visibility → Elasticity → Platform maturity There is no architecture that wins every category. But the trade-offs become much clearer when you evaluate personalization as a system that many development teams will have to build, operate, and evolve. See how the five architectures compare: https://lnkd.in/g2en6zum #SoftwareArchitecture #WebPersonalization #EnterpriseArchitecture #PlatformEngineering #WebPerformance

We put Vercel and Harper head-to-head across 474 tests and 843,240 requests. Harper came out 4 to 14x faster on live, personalized reads. To be fair, Vercel's paid instance with their global CDN was 4ms faster than the free Harper cluster for static data. But personalized reads can't be cached, and Vercel's caching superpower quickly becomes a crutch that limits performance. So, it's your call: will you spend more for slow personalization with Vercel, or speed things up with Harper? See all the results: https://lnkd.in/g3KPqqBX

This week on The Business of Open Source I have an episode with Stephen G. about the process of going from closed source to open source at Harper. It's the second episode in the series, the first one was with Ethan Arrowood, who brought a different perspective to the transition. We talked mostly about the advantages of going open source, both expected and unexpected. Stephen expected that going from closed to open source would shortened the company's sales cycle, or at least reduce the amount of involvement from sales engineers during the evaluation phase, because prospects are able to evaluate the product themselves. That has absolutely happened. What he didn't expect was for support costs to go down, because customers can debug for themselves, send in very detailed bug reports when necessary, often having identified exactly where in the code there's a problem. It's also easier for customers to see when the "bug" is actually a problem on their end, and make the changes necessary without ever needing assistance from Harper. And the rise of agentic engineering has also been an unexpected advantage. Being able to direct an agent directly to the codebase rather than to the docs makes the agents much more able to build on top of and in general deploy Harper, so much so that Stephen wonders if open source will be the default for enterprise software in the future. Check out the episode here: 👇 https://lnkd.in/etiRJV-y

There are a million ways to build a product that looks the same from the outside. But what happens underneath — the architecture, indexing, caching, data model, runtime, and execution path — determines how fast, reliable, and scalable that product can actually be. At Harper, we take those decisions seriously. Our goal is to remove as much infrastructure burden from engineering teams as possible by making strong technical decisions deep in the platform. That means developers can build with confidence, without having to stitch together a stack and optimize every layer themselves. A great example of this appears in Harper 5.1, which introduces major improvements to HNSW vector search using INT8 quantization. The result: → ~3x smaller indexes → ~5x better search throughput → p99 latency for vector lookups reduced from ~9s to ~0.5s under load → enabled by default for new HNSW indexes For teams building semantic search, recommendations, RAG systems, AI agents, or any application that depends on fast vector retrieval, these details matter. And with Harper, those details are built into the runtime. Read the full article from Kris Zyp: https://lnkd.in/gxB_VGGn #VectorSearch #AIEngineering #SemanticSearch #RAG #ApplicationPerformance #Harper

This is what a unified runtime looks like. Database, API, logic, and host in one Harper component, one process. Wild Willows shows it off, but the shape fits any app. Beautiful work, Bailey Dunning. #UnifiedRuntime #AgenticEngineering #DeveloperTools #BuildInPublic

A coding agent can look brilliant in a sandbox. But production systems are not sandboxes. They have history, customers, security rules, dependencies, and a real blast radius. Stephen G. captures that distinction beautifully in this piece, making a complex architectural idea feel clear, practical, and urgent. His core point is that agentic engineering only scales when agents operate within an opinionated architecture. That means: • Shared context • Narrow scope • Deterministic guardrails • Clear system boundaries • A runtime agents can reason inside That is the difference between “vibe coding” and real agentic engineering. Give agents structure, and they stop improvising. Give them Harper, and agentic engineering starts to scale. https://lnkd.in/gD7bCP-V #AgenticEngineering #AI #SoftwareEngineering #Architecture #Harper

Going open source means a few big calls you basically can't take back. Ethan sat down with Emily Omier to walk through how Harper made the move: the licensing, the timeline, where we'd do it differently.

AI is moving fast. Harper 5.1 makes it easier to move with it. In his latest post, Kris Zyp breaks down the new AI stack in Harper 5.1, including: ✅ A provider-agnostic Models API for OpenAI, Anthropic, Bedrock, and Ollama ✅ Built-in agent loops with toolMode: 'auto' ✅ Hard budget controls for cost, tokens, and tool iterations ✅ Parallel tool execution for lower-latency agent workflows ✅ An opt-in Harper Agent component for operator-level tasks like reading logs, deploying components, and adjusting configuration The big idea: AI shouldn’t feel bolted onto your application stack. With Harper 5.1, models, tools, data, application logic, and operational visibility can work together inside the same unified runtime. That means developers can build AI features like summarization, classification, RAG, agentic workflows, and operational agents with less infrastructure to wire together. If you haven’t built an AI application with Harper yet, now is the best time to jump in. Read Kris’s full post here: https://lnkd.in/g6v77GA4

This is an incredible achievement for the team at Harper. We are beyond honored to be named alongside other technologies that will be shaping the agentic engineering era. Really proud of all the work the team did to raise this awareness.