Accessible Agent Interfaces for Keyboard, Screen Reader, and Voice Users
Design streaming conversations, tool progress, approvals, errors, and voice controls that preserve focus, user control, and understandable status.
Stanley Yang — Writing
Stanley Yang turns complex workflows into clear, dependable full-stack products built for real-world demand.
Latest posts
Adopting OpenTelemetry GenAI Semantic Conventions Without Breaking Observability
A migration playbook for adopting the evolving OpenTelemetry GenAI span, metric, and event schema safely.
Agent Swarms: When Multiple AI Agents Help—and When They Make Things Worse
A practical decision framework for choosing a multi-agent architecture, plus a minimal coordinator that keeps concurrency, budgets, and failure handling explicit.
AI Agent Evals in CI: A Release Gate for Prompts, Models, and Tools
Build a provider-neutral agent evaluation suite with deterministic checks, repeated trials, safety cases, uncertainty, and a CI policy that catches regressions.
AI Agent Memory Poisoning: Threat Model, Detection, and Recovery
Protect persistent agent memory with source trust, write gates, quarantine, contradiction checks, provenance, and complete recovery procedures.
AI Agent Memory with Postgres and pgvector: Schema, Retrieval, and Forgetting
Design tenant-safe agent memory with provenance, hybrid search, retention, deletion, and pgvector indexing that does not mistake every transcript for knowledge.
Automatic Prefix Caching for Agent Workloads
Make repeated agent prompts cacheable in vLLM without crossing tenant boundaries, and measure whether cache hits actually improve latency.
Backpressure for AI Agent Systems
Design admission control, bounded queues, concurrency limits, deadlines, and graceful degradation for agent workloads.
Benchmark PostgreSQL 18 Async I/O for RAG Workloads
A reproducible protocol for testing PostgreSQL 18 AIO across vector search, filtered retrieval, ingestion, vacuum, and cache states.
Benchmark Quantized Agent Models for Quality, Memory, and Speed
A reproducible protocol for comparing weight, activation, and KV-cache quantization without inventing savings or hiding quality regressions.
Benchmark Speculative Decoding Reproducibly
Measure speculative decoding against a target-only baseline across load, prompt types, acceptance rates, memory, latency, throughput, and output equivalence.
Benchmark Structured-Output Reliability in Open Models
A reproducible protocol for measuring JSON validity, schema compliance, semantic accuracy, latency, and failure recovery across open models.
Build a Budget Controller for Agent Workflows
Enforce token, tool-call, latency, concurrency, and monetary ceilings across nested agent work without relying on prompt instructions.
Build a Citation-Aware Document Agent for PDFs, Scans, Tables, and Images
Preserve page regions, reading order, tables, OCR alternatives, and source versions so every document answer can resolve to visible evidence.
Build a Citation-First RAG System That Rejects Unsupported Claims
Represent evidence at span level, require citations for verifiable claims, and abstain when retrieval cannot support a grounded answer.
Build a Coding-Agent Harness with Hard Execution Boundaries
Wrap any coding model in a reproducible harness that controls workspaces, commands, credentials, budgets, evidence, and verification.
Build a GPU Capacity Planner for LLM Inference
Estimate weights, KV cache, runtime overhead, context concurrency, and replica demand—then replace assumptions with measured service curves.
Build a Multi-Tenant MCP Gateway Without Collapsing Trust Boundaries
Route remote MCP traffic by authenticated tenant while isolating tokens, sessions, discovery, quotas, results, and audit records.
Build a Plan–Execute–Review Agent Without Infinite Loops
Implement an agent loop with typed plans, evidence-based review, bounded revision, explicit stop states, and safe tool execution.
Build a Production RAG Ingestion Pipeline: Parse, Version, Embed, and Delete Correctly
Design a restartable RAG ingestion pipeline that preserves provenance, detects changes, avoids duplicate work, and removes stale chunks from retrieval.
Build a Real-Time Voice Agent over WebRTC
Design a reconnectable browser voice agent with explicit signaling, ephemeral authorization, audio lifecycle, interruption, and observable session state.
Build a Safe MCP Server in TypeScript: Tools, Validation, and Transports
Build and test a real MCP stdio server, then make the transport, validation, error, and authorization decisions required for a remote deployment.
Build a Simulated World for AI Agent Testing
Create resettable, observable email, calendar, browser, and database environments that test agent effects without touching real systems.
Build a Single-GPU RAG Stack
Reserve one GPU for generation, run embeddings and retrieval on CPU, store cited chunks in Postgres with pgvector, and evaluate retrieval before answers.
Build a Visual Browser-Agent Benchmark That Measures Real Task Success
Create resettable visual web tasks with programmatic outcome validators, layout perturbations, safety checks, and reproducible multimodal traces.
Build AI Agent Tools as WebAssembly Components
Define typed WIT contracts, explicit capabilities, resource limits, provenance, and conformance tests for portable agent tools.
Build an Evidence-First Code Review Agent
Design a review agent that reports reproducible defects with precise evidence instead of producing noisy summaries and style opinions.
Build an Interactive MCP App with a Sandboxed Tool UI
Design an MCP tool that serves an interactive UI resource, receives structured results, and stays safe across hosts with different extension support.
Build an MCP Client in TypeScript: Discovery, Tool Calls, and Trust Boundaries
Connect to a local MCP server, verify capability discovery, call a typed tool, and handle remote transports without confusing discovery with authorization.
Build an OWASP Agent Security Test Harness
Turn the OWASP Top 10 for Agentic Applications into safe, repeatable abuse cases with simulated tools, hard invariants, evidence, and CI gates.
Build Forensic Audit Logs for AI Agents
Capture actor chains, policy decisions, approvals, tool effects, and evidence so investigators can reconstruct an agent incident without logging hidden reasoning or secrets.
Build Long-Running MCP Tools with the Experimental Tasks Utility
Design a durable MCP tool call with polling, deferred results, input-required handling, cancellation, and tenant-safe retention.
Build Streaming Agent Interfaces in Next.js
Stream agent UI and typed run events with Suspense, Route Handlers, cancellation, backpressure, and accessible status updates.
Build Two Agents That Collaborate over A2A 1.0
Connect a coordinator to an independently operated report agent using an Agent Card, SendMessage, task state, typed artifacts, and strict authorization.
Caching AI Applications: Prompts, Prefixes, Retrieval, and Tools
A correctness-first guide to caching model prefixes, retrieval results, tool calls, and final AI outputs safely.
Capability-Based Security for AI Agent Tools
Replace broad ambient credentials with narrow, expiring, attenuable grants bound to one agent task, tenant, operation, and resource set.
Choose an Open Model with an Evaluation Suite
Select open weights with a reproducible, product-specific suite covering capability, tool use, safety, latency, memory, and operational fit.
Computer Use vs DOM Automation vs APIs: Choose the Most Reliable Agent Interface
Choose between APIs, structured browser automation, accessibility trees, and visual computer use using reliability, coverage, security, and cost.
Context Engineering for AI Agents: Build a Context Assembler
Design a context pipeline that selects trusted instructions, relevant evidence, tool definitions, and recent state within an explicit token budget.
Contract Testing for AI Agent Tools
Test agent tool schemas, transport behavior, authorization, retries, and compatibility with valid, invalid, and adversarial requests.
Cost Attribution for Multi-Agent Systems
A practical ledger for attributing model, tool, compute, retry, and shared costs to successful multi-agent outcomes.
Delegated Authorization for AI Agents Acting on Behalf of Users
Preserve user intent and actor identity across agent-to-service and agent-to-agent calls with narrow audiences, scopes, consent, and delegation chains.
Design an Async Sandboxed Agent Tool Runner with WASI 0.3
Use ratified WASI 0.3 async components with pinned runtime support, explicit capabilities, deadlines, backpressure, and safe cancellation.
Design Approval UX for High-Risk Agent Actions
Turn human approval into a real security control by previewing exact effects, binding approval to immutable action data, and preventing replay or substitution.
Design-to-Code with Visual Regression That Developers Can Trust
Turn design states into deterministic Playwright checks that combine semantic assertions, visual baselines, accessibility, and human-reviewed diffs.
Deterministic Replay for Nondeterministic AI Agents
Record model, tool, time, randomness, configuration, and state dependencies so agent failures can be replayed without repeating real-world effects.
Disaggregate Prefill and Decode for Long-Context Agents
Separate prefill and decoding only when independent scaling and latency isolation beat KV-transfer cost, duplicated weights, and operational complexity.
Do Repository Instruction Files Help Coding Agents? A Benchmark Protocol
Measure absent, concise human-written, generated, and noisy repository instructions with paired tasks, fixed agent settings, and no invented results.
Durable AI Agents: Retries, Idempotency, Human Approval, and Resume
Design long-running agent workflows that survive crashes, retry only safe work, bind approvals to exact actions, and reconcile ambiguous side effects.
Edge vs. Node vs. GPU Backends for AI Applications
Choose AI backend placement with a reproducible latency, quality, capacity, cost, security, and failure benchmark.