The Levr Blog
Building software with humans and agents.
Field notes on AI-native project management, agentic QA, and building software with humans and agents on the same team.
Why AI agents need wallets for micropayments
An AI agent wallet gives a coding agent a controlled way to pay for APIs, data, communications, and compute without stopping for a human checkout flow. It matters when an agent needs a small, immediate purchase to finish work, such as purchasing premium research, sending a message through a paid ser
Why Cursor teams choose Levr over Linear agents
While Linear's agent integrations ease admin work, human-centric trackers lack a native quality loop. Coding agents need more than status updates, they require automated test tracking. See why Cursor and Claude Code teams are moving to Levr's agent-first project control plane.
x402 payments for AI agents on AWS
x402 payments are a machine-to-machine payment pattern built around the HTTP 402 Payment Required status code. Instead of forcing an autonomous agent through a human subscription, API-key setup, or card checkout, a service can return a payment requirement and let the agent authorize settlement befor
Verification debt: how to verify AI-generated code
How verification debt builds up when AI-generated code outpaces review capacity — zero-trust, multi-layered checks, and CI/CD quality gates that catch it before it compounds into rework. The goal isn't slowing agents down, it's making their speed trustworthy.
How the Codex harness works for coding agents
How the open-source Codex harness manages context, tool execution, sandboxing, and long-running agent workflows — architectural patterns for building reliable, fast coding agents beyond one implementation
AI workflow automation for coding agent teams
How AI workflow automation turns one-off coding prompts into repeatable, secure playbooks — deterministic permissions, shared plans, and verification loops for dependency upgrades, CI triage, and pull request prep.
Agent harness: the foundation of AI-native software
AI-native software is software designed around models that reason, call tools, maintain context, and adapt interfaces while work is underway. Rather than attaching a chat box to an existing application, it treats agent coordination, multimodal input, long-running tasks, and human approval as core pr
Multiplayer agentic engineering for software teams
Multiplayer agentic engineering is a software development approach where people and coding agents collaborate on the same work across shared tools, sessions, project context, and review processes. Rather than leaving an agent isolated in one developer's terminal, the team can inspect its reasoning,
How GitHub merge queues prevent AI velocity sickness
How GitHub merge queues and stacked pull requests help AI-powered engineering teams stay coordinated to prevent the PR overload, merge friction, and lost context that come with AI-accelerated code output.
Coding agent benchmarks for legacy codebase refactors
A practical benchmarking framework for coding agents tackling legacy codebase refactors. Evaluation contracts, success-rate thresholds, and scoring rubrics that measure real completion, not just fast output.
How to review and secure AI agent skills
AI agent skills are reusable instruction packages that teach coding agents how to perform a specific task reliably, such as configuring an API security scan, deploying a service, or creating a pull request. A well-built skill gives an agent clear activation criteria, safe operating boundaries, and r
Claude Certified Architect exam guide for agentic engineers
A practical roadmap through the Claude Certified Architect exam's core domains — agent loops, tool use, Claude Code repository structure, MCP design, and context compaction — built for engineers shipping reliable agentic systems, not just passing a test.
Software supply chain security for AI coding agents
How AI-generated contributions and supply chain risk are straining open source — and why verification, least-privilege access, and open-weight flexibility matter more than ever for coding agent teams.
Compliance automation software for AI coding agents
How SOC 2 controls, segregation of duties, and compliance automation software keep AI coding agents auditable and safe to deploy in regulated fintech and energy trading environments.
How to stop your AI second brain from rotting
An AI second brain is a knowledge system that collects notes, conversations, project material, and operational facts so a coding agent can retrieve useful context later. Its central maintenance problem is knowledge decay: information that was accurate when captured can become outdated, duplicated, o
AI agent sandboxing: a zero-trust architecture
A technical blueprint for securing autonomous AI agents through zero-trust architecture, cryptographic workload identity, microVM isolation, and network-level containment that make agent compromise inconsequential rather than catastrophic.
Open weight models: trust, control, and optimization
Why open-weight models give engineering teams real control and inspectable weights, post-training customization, and predictable costs — plus what Kimi K3's release reveals about the current open-weight landscape and its licensing tradeoffs
How NVFP4 compresses LLMs without breaking them
A practical look at how NVFP4 and 4-bit quantization compress large models without breaking instructions, long-context handling, or coding reliability and, how to validate a compressed model before trusting it in production.
Model routing for AI coding agents
Model routing is the practice of selecting, coordinating, and switching between language models as an AI system works. Rather than sending every prompt to one expensive frontier model or one low-cost model, a router assigns work according to task difficulty, model strengths...
Where does agent context live?
Agent context is the information a coding agent uses to understand its goal, constraints, relevant systems, and evidence of success. It includes written artifacts such as issues, docs, code, and test results, but it also includes decisions and operational knowledge...
Claude Code context management: what to keep or cut
AI layer ablation is a controlled evaluation that removes instructions, skills, hooks, or delegated agent configurations to measure whether they still improve a coding agent’s results. For teams using Claude Code, the goal is not to blindly erase working project knowledge.
How iframe sandboxing secures personal AI-generated apps
Personal AI codegen is the idea that people can use coding agents to adapt software for their own needs instead of waiting for a vendor to prioritize every request. That model challenges conventional cloud delivery, where one centrally hosted application serves every customer...
Claude Code skills: a practical workflow guide
AI coding agent skills are reusable instruction packages that give coding agents repeatable ways to plan, clarify, configure, and verify work. Rather than packing every rule into a long project prompt, teams can invoke a focused skill only when a task needs it. A useful skill library also needs docu
Advanced MCP App Patterns: Bidirectional State and Reusable Views
MCP Apps let a Model Context Protocol server return an interactive interface instead of a long text reply, preserving visual context and supporting real actions inside a sandboxed, host-controlled environment.
Designing Test Environments for Long-Horizon AI Agents
Long-horizon AI agent evaluation measures whether a coding agent can complete work requiring sustained reasoning, tool use, and recovery from earlier mistakes — not just how many tokens it consumes or how long a task takes.
How to Evaluate AI-Powered Code Review
Agentic code review uses coding agents to inspect pull requests, apply repository-specific rules, and surface risks for human engineers before merge. It is most useful when teams need broader coverage without treating model output as authority.
Why AI Coding Agent Context Rots and How to Fix It
AI coding agent context is the instructions, repository guidance, skills, and tool information supplied to a coding agent beyond its base model knowledge. It can help an agent follow local conventions, find platform-specific workflows, and use the right commands.
When Can AI Agents Merge Their Own Code Safely?
Harness engineering is the discipline of building the context, rules, checks, review paths, and feedback loops that let coding agents produce useful software without degrading a codebase over time. It shifts engineering attention from writing every line manually ...
When should AI agents require human approval?
Human approval for AI agents is the practice of requiring a person to review or authorize an agent’s proposed action before it can affect another person, system, or business outcome. It matters most when an action is difficult to undo, uses sensitive context, or reaches...
Vector Search and Failure Testing for AI Infrastructure
Building AI infrastructure means turning a useful systems insight into software that remains economical, reliable, and operable as demand grows. The hard part is rarely choosing a fashionable database or adding more hardware. It is understanding the workload, ...
Tool Binding and Runtime Control in AI Coding Agents
Agent harnesses are the operational layer around a model that gives a coding agent tools, safety limits, memory, feedback loops, and execution control. In practice, the harness often determines whether an agent can complete real work, verify its output, and stay within safe boundaries.
How Engineering Managers Measure AI Coding ROI
Software factory is a practical term for a development system where agents do more than assist with code completion. They pick up scoped work, generate changes, review output, file follow-up tasks, and push delivery forward with limited human intervention.
Escaping the Shipping Trap and the Research Trap in AI Coding
Coding agent adoption often stalls in a predictable way. Teams either keep shipping around agent mistakes, which preserves short-term velocity but prevents real improvement, or they pause delivery to invest in autonomy, which can drag productivity down for months. The result is a ...
AI Engineering's New Bottleneck: Human Attention, Not Code
AI engineering is shifting from writing every line of code by hand to designing systems where coding agents can plan, build, test, and hand work back for review. That shift matters because modern agents are no longer limited to autocomplete. They can handle longer tasks, run tools, and ...
How to build product teams in the age of AI
AI-assisted product development means software teams use coding models and agentic workflows to turn ideas into shipped changes faster than ever, while still keeping humans accountable for system design, safety, and adoption. The hard part is no longer just writing code. It is deciding what to build
How to make coding agents reliable in judgment-heavy workflows
Coding agents can automate complex work, but in judgment-heavy domains, a correct-looking answer is not enough. You also need confidence that the process used to reach that answer was valid, traceable, and repeatable. That is the core challenge when agents work across large, messy datasets, especial
Building an AI-Powered Knowledge Base for Production Agents
Karpathy LLM Wiki style agent memory works extremely well when the agent is just for you: a local folder of markdown, fast iteration, and flexible knowledge you can shape over time. But once you ship that agent to other users, the architecture has to change.
Teaching AI Agents Better Retrieval: Beyond Keyword Search
Agent retrieval is the process a coding or research agent uses to find the right information before it reasons, writes, or acts. When retrieval is weak, even strong models miss crucial evidence, rank the wrong sources, or answer with partial context.
Designing AI Interfaces That Fight Automation Bias
Human-in-the-loop AI is not automatically safer just because a person approves the output. In many systems, humans gradually stop evaluating and start confirming, especially when the interface frames AI output as likely correct. That creates automation bias, weaker decisions, and ...
Scaling Coding Agents: What Agent Enablement Actually Means
Agent enablement is the set of team, platform, and governance practices that help coding agents work reliably across an engineering organization, not just on one developer’s laptop. It covers reusable context, clearer task definitions, evaluation, observability, access control, and ...
Pydantic AI 2.0: How capabilities change AI agent design
Pydantic AI 2.0 is a major shift in how you build production agents. Instead of wiring instructions, tools, hooks, and guardrails into one messy constructor, you package each responsibility as a reusable capability. That makes agents easier to compose, easier to maintain, and easier to share across
Reinventing the Dev Team for Agentic Software
Agentic software development is changing how engineering teams plan, build, and ship software. Instead of treating code generation as a small productivity boost, teams now face a bigger operational shift: development can move faster than product decisions, testing, and release processes can keep up.
The Product Brain: How PMs Can Orchestrate AI Coding Agents
Product brain refers to a structured system that continuously gathers product context, synthesizes it into usable knowledge, and helps agents act with better judgment. Instead of relying on scattered docs, tickets, and chat history...
The Missing Layer After Launching AI Agents
Production AI agents are not finished when they ship. They become an operational system that needs monitoring, diagnosis, evaluation, and continuous improvement across real conversations. A demo can look solid while the live product quietly drops constraints, chooses the wrong tool, or reaches techn
Adaptive engineering beyond fixed AI harnesses
Adaptive engineering is the idea that an AI system’s operating structure should not always be fixed before work begins. Instead, the harness, meaning the prompts, roles, tools, sequencing, and memory around a model, can reorganize during execution as the problem changes. This matters most when codin
Why Autonomy Isn't the Goal: Finding the Right AI Coding Setup
AI coding levels describe how much control you hand to a coding agent, from simple autocomplete to fully autonomous software delivery. The useful question is not how autonomous your setup looks. It is how reliably it turns ideas into correct, maintainable code. For most teams, the sweet spot is not
How to embed a coding agent in your product
Embedding a coding agent in your product means giving software an adaptive runtime that can inspect context, choose from available tools, and complete tasks without a rigid step-by-step workflow. Instead of hard-coding every branch, you define instructions, expose the right capabilities, and...
What's New in Skills v1.1: A Guide to Wayfinder and the Updated Workflow
Skills v1.1 is an update to an AI coding workflow that turns scattered prompting into a more structured software delivery system. The release adds a clearer planning flow, new implementation and research skills, and a better way to handle projects that are too large for a single agent session.
What is verifiable continual learning for AI agents?
Verifiable continual learning is a way to help AI agents improve from real failures without breaking skills that already work. Instead of treating production logs as enough on their own, it turns failures, traces, and feedback into replayable learning environments with clear success criteria. That m
Browser-Native AI Agents: Running Locally, Safely
Browser AI agents are coding or automation agents whose planning loop, tools, and execution live inside the browser itself, not in a remote container or cloud sandbox. Instead of treating the browser as a thin chat surface, this model uses the browser as runtime, isolation boundary, and control surf
How do you benchmark agentic AI workloads?
Agentic AI benchmarking measures how systems perform when models operate across many turns, call tools, reuse context, and build long working histories instead of answering a single prompt once. Unlike classic LLM benchmarks, it must account for cache behavior, routing, tool latency, token generatio
SWE-Marathon and long-horizon coding agent evaluation
SWE-Marathon is a benchmark for testing whether coding agents can handle project-scale software work over extremely long runs, not just short bug fixes or isolated functions. It extends the benchmark lineage from HumanEval and SWE-bench into environments where agents must plan, edit, test, recover,
Why every coding agent needs its own desktop
A coding agent desktop is an isolated development environment, usually a full remote workspace with an IDE, browser, tools, and repo state, assigned to one agent for one task. Instead of several agents colliding inside the same checkout, each agent gets its own machine context, can plan, implement,
What Is Tessl Agent and How Does It Improve Coding Agents?
Tessl Agent is a system designed to make coding agents better over time by finding repeated mistakes, recurring chores, and workflow gaps across pull requests, issue trackers, and agent session history. Instead of relying on more manual prompting or one-off harness tweaks, it identifies patterns and
Why you shouldn't optimize AI agent costs manually
AI agent cost optimization is the practice of reducing spend from coding agents, model calls, and automated software workflows without hurting output quality or slowing delivery. In most teams, the biggest mistake is asking developers to manually choose the cheapest model for each task. That sounds
The Pipeline Is Dead: What Replaces It?
Adaptive software is software that can change per user, per context, and even during runtime instead of being frozen once and shipped identically to everyone. That challenges the core assumption behind CI pipelines, package registries, container images, and release workflows: one build, one artifact
What Is Claude Tag and How Does It Work?
Claude Tag is a proactive coding agent that works inside team chat, keeps context across channels, and can carry work forward over longer periods instead of waiting for one prompt at a time. Unlike a standard coding assistant session, it can follow team workflows, surface what needs attention, and h
What Is Open Knowledge Format for LLM Wikis?
Open Knowledge Format, or OKF, is a lightweight standard for building markdown-based knowledge bases that language models can read, search, and maintain consistently. It extends the popular LLM wiki pattern into a shareable structure with predictable folders, index files, and metadata, so one agent
What are MCP Apps and how do they work?
MCP Apps are interactive applications built on top of the Model Context Protocol that let a host such as Claude, ChatGPT, Codex, or Cursor render UI from an MCP server instead of showing only plain tool output. Unlike a basic MCP server that returns structured data or text, an MCP App can return a s
Harness engineering and the path to a software factory
Harness engineering is the practice of building the feedback loops, checks, and workflows that let coding agents do more of your software delivery work with less human correction. Instead of treating agents as one-off assistants, you treat them as parts of a system that can be improved over time. Th
What Is Agentic Software Development?
While agentic software development is rising, traditional human-centric tools bottleneck autonomous AI. To scale, teams need a dedicated control plane, a centralized system providing agents with shared context, live memory, and human-in-the-loop governance.
Designing AI Interfaces Beyond the Prompt Box
Prompting is still the dominant way people use modern AI, but it remains structurally similar to older batch-style computing. You package intent, submit it, wait, inspect the result, then try again if needed. That workflow feels natural only because it has become familiar, not because it is the best
The shipping sprint that exposed the real AI bottleneck
Vibe coding inside a large company is not just fast prototyping with AI. It is a temporary shift in how teams make decisions, share context, and move from idea to shipped software with fewer handoffs. In a month-long experiment at Automattic, small autonomous teams paused normal roadmap work and bui
Accept the risk, sandbox everything for AI security
Prompt injection is the reality that language-driven AI systems can be manipulated through inputs you cannot fully predict or sanitize. For teams building with coding agents, copilots, or internal AI tools, the practical response is not total prevention. It is containment. That means treating AI as
Claude Code Workflow: Give Your Agent a Project, Not a Prompt
A Claude Code workflow defines the infrastructure and context boundaries surrounding Anthropic's terminal-based AI coding assistant. While the model excels at generating and refactoring code within isolated sessions...
How malicious SKILL.md files can compromise AI coding agents
SKILL.md security is the practice of treating agent skills like software supply chain components, not harmless markdown snippets. A skill can contain instructions, references, downloads, and behaviors that push a coding agent to fetch files, read secrets, install packages, or send data elsewhere.
Why skills are the new code in agentic development
Skills in agentic software development are reusable pieces of context that tell coding agents how to act, what standards to follow, and how to complete recurring work safely and consistently. As development shifts from hand-writing every implementation detail to expressing intent, skills increasingl
What AI coding did to dev teams in 6 months
AI coding is changing software teams by shifting the hardest work away from typing code and toward specification, architectural judgment, verification, and operational supervision. Teams using tools like Claude Code, Cursor, and similar assistants often find that code arrives faster
Agent experience: redesigning dev platforms for AI agents
Agent experience, or AX, is the practice of designing software platforms so humans and coding agents can work together safely and efficiently. It goes beyond API compatibility. AX covers structured errors, machine-readable signals, event-driven workflows, clear system context, and approval paths...
Local LLMs: Your Fallback when Frontier Models Get Restricted
Local models are AI models that run entirely on your own hardware instead of a provider's servers. That means no API dependency, no per-token billing, and no risk that a policy change or outage suddenly removes access to a critical workflow.
WebMCP: making web apps faster and cheaper for coding agents
WebMCP is an emerging browser API that lets a web app expose structured tools to coding agents instead of forcing them to interpret screenshots, scrape the DOM, or guess selectors. In practice, that means agents can interact with the front end through an explicit contract defined by the page author,
The production AI playbook for enterprise agent deployment
Production AI transforms agent demos into resilient, measurable enterprise systems. Rather than starting with model selection, success requires defining business outcomes, building evaluation data, and tracing decisions before launch
Human in the loop: what to delegate vs. keep reviewing
Today, AI-native development is software delivery where coding agents handle much of the implementation, while humans define requirements, guide architecture, and review results. Like Level 4 autonomous driving, humans set the destination while staying ready to take control when needed.
Executable specs for agentic coding: how to build a verification layer
Executable specs turn product requirements, design constraints, and runtime behavior into machine-checkable inputs that coding agents can verify before code is merged. Instead of trusting generated code because it compiles, teams validate whether a feature matches intent, works in a real preview env
Securing AI Agents in Production: Prompt Injection, Sandbox, Supply Chain
AI security now extends beyond model misuse into software supply chains, prompt injection, sandboxing, least-privilege access, and the design of web systems that agents can safely operate. As coding agents move from autocomplete to execution, teams need better visibility, tighter controls, and...
tldraw and AI Agents: A Visual Approach to Coding
Agents on the canvas describes a way of using an infinite whiteboard as a shared workspace where humans and coding agents can see, interpret, and modify the same visual context. Instead of keeping AI inside a chat box, tools such as agent-first development platforms and canvas systems like tldraw tu