AI Engineering
47 essays tagged with AI Engineering.
Jev and System One Models: What Changes When AI Returns Decisions?
A critical engineering review of Jev and System One models: typed probabilistic decisions, calibration, abstention, latency, schema design, governed execution, and the evidence required before automation.
Agentic Inference: A Practical Guide to KV Cache, Scheduling, and Throughput
Why tool-using agents turn inference into a stateful memory-scheduling problem, and how to measure, design, benchmark, and roll out a better serving path.
Harness Engineering: The September 2026 Practitioner’s Field Guide
A primary-source review through September 15, 2026: controlled harness comparisons, evolving safety defenses, context inheritance, managed runtimes, implementation availability, and the next experiments.
Managed Agent Harnesses: The Control Boundaries You Still Own
A research-backed operating guide to managed agent harnesses: durable sessions, caller credentials, authorization, interrupted effects, version drift, migration tests, and application-owned evidence.
FrontierHarness Eval: Auditing the 17.5× Cost Gap
A reproducible audit of FrontierHarness Eval: cost per success versus median cost, paired task outcomes, small-sample uncertainty, and what to measure before choosing a production harness.
A Research and Publishing Roadmap for Harness Engineering
A prioritized roadmap for original harness engineering research and technical publishing, with experiments, baselines, artifacts, decision gates, and falsifiable directions through 2027.
Harness Engineering in September 2026: Measurement, Evolution, and Operational Reality
A critical research map of harness engineering through September 10, 2026: new benchmarks, automatic optimization, reusable tools, long-running agents, public implementations, and unresolved questions.
Agent Harness Benchmark: An Experimental Protocol Beyond Model Scores
A proposed benchmark protocol for isolating harness effects, injecting operational faults, measuring repeated reliability, and publishing auditable scorecards.
From Feature Stores to Agent Memory: What Production ML Teaches Agent Systems
Production ML lessons for agent memory: availability-time correctness, historical replay, versioned materialization, promotion, correction, and deletion across derived stores.
Welcome, Astra: What GPT-6 Changes About Intelligence at Work
A research analysis of OpenAI's GPT-6 Astra: benchmark evidence, scientific discovery, tool use, safety, economics, and an evaluation plan for real work.
The AI Agent Review Tax: When Automation Creates Verification Debt
Measure the review, repair, and integration work behind agent output—and design verification that catches errors without erasing the gain.
Proof-Carrying Context: Why AI Agents Need More Than a Context Window
A research-grounded revision of context engineering: from relevant tokens to a replayable view of evidence, conflicts, omissions, and decision sufficiency.
Agentic Recommender Systems: How Harness Engineering Changes Personalization
Recommenders are evolving from rankers into tool-using agents. Here is the hybrid architecture, evaluation harness, and rollout plan needed to build them safely.
Prompt Engineering and LLM Latency: The 10-Day Field Report
A source-backed field report on the last ten days of prompt engineering and LLM latency: OpenAI Ultrafast and long-context Fast mode, vLLM and SGLang releases, new inference research, production best practices, and the roadmap ahead.
Production Prompt Engineering in 2026: From Instructions to an Evaluated Contract
A deep production guide to prompt engineering: lean instruction design, stable cacheable prefixes, tool and output contracts, reasoning budgets, datasets, robustness tests, versioning, rollout gates, and a 90-day implementation roadmap.
Persistent Memory Poisoning: The Attack That Outlives the Session
A production security architecture for attributable, revalidated, authority-bounded, traceable, and selectively reversible agent memory.
Multi-Agent Consensus Is Not Correctness: How Debate Manufactures Confidence
Why agent agreement can hide correlated error—and how to build governed deliberation with effective agent count, independent verification, and release gates.
Agent Evaluation After the Leaderboard: The KDD 2026 Production Assurance Playbook
KDD 2026 shows why agent evaluation must cover protocols, trajectories, external state, evaluator operating profiles, causal rollout, and drift.
OpenAI's Latest Work: A 10-Day Field Report
A source-backed 10-day delta across Codex, plugins, API operations, model economics, security scanning, and the GPT-5.4 migration.
OpenWorker Review: A Real Desktop Coworker With an Unfinished Trust Runtime
OpenWorker already owns the agent loop, approvals, connectors, and desktop UX. Its next leap is containment, durable effects, replay, budgets, and evals.
Anthropic in 2026: A Full-Stack Map of Its Latest Work
A source-backed map of Anthropic's latest work across Claude models, agents, the developer platform, safety, research, enterprise strategy, and compute.
Inside Anthropic's 2026 Claude Platform: Managed Agents, Tools, Memory, and MCP
A builder's map of the Messages API, Agent SDK, Managed Agents, programmatic tools, MCP, Skills, memory, task budgets, and enterprise controls.
Claude 5 Field Guide: Fable, Mythos, Opus, Sonnet, and Haiku
A practical guide to Anthropic's current Claude lineup, including capability tiers, pricing, context, effort, safeguards, fallbacks, and model-selection tradeoffs.
GPT-5.6 Field Guide: Models, Tools, Reasoning, and the New Runtime
A practical guide to GPT-5.6 Sol, Terra, and Luna—and the runtime features that matter more than a headline benchmark score.
OpenAI in 2026: A Full-Stack Map of Its Latest Work
A source-backed map of OpenAI's newest work across models, agents, multimodal products, science, safety, and compute—and how the pieces fit.
OpenAI's Multimodal Stack: GPT-Live, GPT Image 2, and Sora 2
A product and developer guide to OpenAI's latest voice, image, and video systems—and the boundaries between interaction and media production.
Inside OpenAI's 2026 Agent Platform: Responses, Tools, MCP, and Control
A builder's map of the Responses API, programmatic tools, multi-agent orchestration, MCP, identity, spend controls, and deployment boundaries.
OpenAI's Latest Science Work: From Benchmarks to Verifiable Discovery
What OpenAI's newest mathematics, scientific-computing, biology, and evaluation work says about AI as a research collaborator.
Graph Engineering for AI Agents: The X Hype Cycle and a Production Playbook
Separate graph-engineering signal from X hype, then compile agent workflows with governed transitions, graph invariants, durable effects, and exact replay.
Reverse Prompting: Work Backward from the Output, Then Prove the Prompt
Reverse prompting is useful when it turns a desired output into a candidate instruction contract and validates that contract forward. Here is the research, the workflow, the failure modes, and the security boundary.
Agent Memory Should Be Compiled at Recall, Not Replayed From Storage
What MemHarness teaches us about reconstruction, negative transfer, source state, and compiling production agent memory against current evidence, policy, and authority.
Human Oversight for Agent Fleets: Confidence Is Not an Audit Policy
A new audit-allocation paper shows that self-reported confidence can make limited human review worse than random and that tiny audit budgets can become rubber-stamping. Production oversight needs risk gates, stratified random coverage, correlation-aware learning, and a measured non-vacuity test.
Passive Awareness Is the Missing Multi-Agent Primitive—But It Needs a Control Plane
AgentRadio shows that agents improve when discoveries can cross lane boundaries during execution. Production teams should keep the asynchronous channel, then add typed evidence, bounded delivery, authority isolation, and replay.
Adaptive Agent Harnesses: Learn From Experience Without Letting Production Rewrite Itself
MemoHarness shows that execution experience can improve the control layer around an LLM. Production systems need a stricter pattern: adaptive performance inside an immutable safety envelope.
Agent Memory Is a Systems Workload: What SelfMem Changes—and What It Does Not
SelfMem improves long-horizon recall by letting an agent optimize its memory strategy. New systems and security research shows the production contract must also cover cost, freshness, provenance, and poisoning.
Your Agent Benchmark Is Measuring the Harness
Two new evaluation systems separate benchmark, harness, model, and environment. Their results show why one leaderboard score cannot tell you which agent to ship.
Multi-Agent Systems Need Replay Before More Agents
Multi-agent systems multiply coordination, cost, and causal ambiguity. Prove one agent is replayable before adding lanes and handoffs.
Threat-Model an AI Agent: Sources, Sinks, Authority, and Blast Radius
A practical AI agent threat-modeling method that maps untrusted sources to dangerous sinks, then constrains identity, authority, data, and blast radius at deterministic runtime boundaries.
The AI Software Delivery Squad: From Ticket to Proof-Carrying Pull Request
A production blueprint for coding agents that scope, patch, test, review, and open pull requests without inheriting merge or deploy authority.
The State of AI Agents in 2026: Standards Converged, Models Improved, Production Moved to the Harness
A mid-2026 review of agentic AI: MCP, A2A and AP2 converged as standards and models got more reliable — yet the bottleneck moved to the governed agent harness.
AI Agent Memory Is Broken: Designing Multi-Layer Memory for Production AI Agents
A production guide to AI agent memory architecture: designing long-term memory for AI agents across working, episodic, semantic, procedural, and organizational layers. Why RAG is not memory, why vector databases are not memory, and how governed, situation-aware memory prevents memory poisoning in enterprise AI agents.
Agent Harness: An Architectural Framework for Production AI Agents
A whitepaper on typed contracts, policy gates, traces, verification loops, and release control for production AI agents.
Dataset-First Agent Engineering: The Golden Sets Behind Reliable Agents
A practical guide to golden sets, task distributions, corrected runs, held-out releases, and production slices for agent engineering.
How Great AI Engineers Build Agents: Datasets, Scores, and Harnesses That Improve
Why strong AI engineers build datasets, scorecards, traces, and improvement loops instead of treating agents as prompts plus tools.
Harness Candidates Are Model Checkpoints: How to Improve Agents Without Silent Mutation
How to treat every prompt, retrieval, tool, policy, and evaluator change as a scored, reviewed, reversible harness candidate.
Scorecards Over Vibes: The Five Metrics That Keep Agents Honest
The five metrics that keep agents honest: policy, utility, latency, safety, and economics.
Trace Review Is the Agent Debugger: Grade the Path, Not Just the Answer
How trace review grades the path, not just the answer, by inspecting context, plans, tools, guardrails, critic verdicts, and corrections.