Agent design patterns

A Stanford and Together AI paper reports 62.5–72.9% less decoding compute with higher accuracy from Hybrid Reasoning Agents. HybridLink switches between latent reasoning and text-based tool calls.

Geoffrey Huntley's experimental SBCL kernel uses LLM tool calls to add functions, test them and hot-reload passing changes. It saves revisions with rollback support; a fresh process reproduced a successful revision.

Developers report using Jev to score inbox content, classify invoices, and split documents into categories. DocJev’s author claims 6x lower latency than GPT-5.6 Luna at equivalent accuracy for document classification and splitting.

TypeSafe says jev-use generates candidate actions from browser state, has Jev select one, then validates and executes it through Cua Driver. Its CUA-S1-FORMS model scored forms locally in 7–9 ms, excluding execution.

Stagehand says adding Jev to its browser primitives cut median Act latency from 1.97 seconds to 0.46 seconds. Jev handles page-level choices and falls back to an LLM when uncertain.

CUA open-sourced CUA-S1-FORMS, a specialist model that selects bounded actions such as filling fields, checking boxes, clicking, or skipping. Cua Driver executes the ordered plan, and the MIT release includes synthetic-data generation, training, evaluation, and deployment tools.

Polylane says it replaced role-specific sub-agents with one main agent and improved quality while reducing latency and cost. The report is a practitioner case study, not a general benchmark.

Nous Research says a 15-hour Hermes Agent run used waves of roughly 120 subagents on one desktop machine to simplify its repository. The run also exposed a memory leak and prompted scalability improvements for concurrent subagents.

Anthropic released blueprints and reference implementations for shopping and merchant agents in four verticals. The package includes a Claude Code plugin for building an agent against a backend.

Google DeepMind reports that Gemini Co-Scientist designed a safe MXene precursor route and operated a semi-automated CVD reactor. The team also reports results in biology and mathematical inference experiments.

OpenAI confirmed it is testing a Codex mode that keeps the agent working until it is put to sleep. Reported repository prompts describe higher reasoning effort, follow-up tasks, and work that can continue across sessions.

Across 60 research projects, ASI-Bench found full procedures averaged 50.91, versus 29.10 for prompts naming only a method. Other evaluations similarly measure whether procedural skills improve execution rather than merely adding more instructions.

Anthropic says Claude ran computational protein-design campaigns from expert-written protocols, producing binders tested by independent labs. Across 1,320 designs, reported hit rates ranged from 22% to 35%.

A study summary reports 307 confirmed failures caused by agent skills, including 125 functional failures. Practitioners favor user-invoked skills to avoid ambiguous automatic triggers and token use before a skill is needed.

Practitioner threads argued coding agents work better with task-specific harnesses, compact runbooks, evals, and deterministic checks. Examples included Sentry MCP traces and state-machine guardrails.

Practitioner posts argued agent evals should check final world state and tool-call trajectories, not just single outputs. A 288-run AGENTS.md test found context files did not improve correctness.

A paper summary found agent skills caused regressions across nearly 6,000 paired office-automation runs. VS Code also added prompt-to-skill migration, making the workflow more accessible despite reliability cautions.

Practitioners are sharing smaller, staged coding-agent loops instead of one large prompt. Examples include a Claude Code software factory, project-specific review agents, searchable logs, and harness layers for evals and observability.

Practitioner threads favored narrow tasks, deterministic steps, task-specific harnesses and explicit acceptance criteria over open-ended agent loops. One Codex workflow reported 58.6%-87.0% token cuts on SWE-bench task sets.

Reports described GPT-5.6 Sol adding needless abstractions and Fable spending quota on many subagents for small changes. The debate frames lighter setups and senior review as safeguards against agent-made tech debt.

Practitioners described GitHub or folder-based markdown knowledge bases that feed persistent company or personal context to Codex, Claude Code, and Hermes. OpenWiki added codebase and personal brain modes for the same pattern.

ClaudeDevs reports Sonnet 5 with a Fable 5 advisor reached ~92% of Fable 5's SWE-bench Pro score at ~63% of the price. Other builders route implementation to Sonnet, Codex, GPT-5.5, or GLM workers.

Fable 5 users reported Opus 4.8 fallbacks, $600 Max-account rotations, slow browser automation, and token-saving subagents. Watch routing opacity, quota burn, and latency before relying on it for long-running agent work.

ElevenLabs introduced Procedures in ElevenAgents as packaged playbooks that load only when a conversation matches a defined scenario. Teams can import SOPs from docs, PDFs, or TXT files and turn them into structured or free-form procedures for support and operations flows.

Apify added more than 20,000 Actors to the x402 flow, letting agents pay in USDC and run tools on demand through HTTP 402 responses. That gives agents a way to buy web automation tasks without pre-provisioned API keys or a manual checkout step, so builders can test paid tool use directly.

A day after /goal and thread automations landed in Codex, practitioners started standardizing on /goal specs, /fork or /side detours, and /rewind plus /compact recovery. The pattern matters because verifier design and compaction timing now control how well long runs hold together.

Microsoft open-sourced SkillOpt, a system that treats agent skill documents as tunable artifacts and improves them against measured task batches. It matters because practitioners are already standardizing shared /research, QA, and packageable skills across harnesses, turning skill files into a new optimization surface alongside models.

Plannotator v0.21.3 shipped file-scoped comments, a unified review UX, default per-file Ask AI chats, and a more reliable Codex app-server path. It matters because guided reviews and plan checks can now plug into agent workflows with less custom glue.

Codex users documented thread automations as recurring wake-up calls that preserve thread context, alongside /goal and /btw patterns for steering long-running loops. The workflow matters because teams can schedule check-ins, queue instructions mid-run, and add adversarial review passes without building a separate orchestrator.

Junior’s first memory system cut one analytics task from about 3 minutes to 1 minute in early tests, with tokens down two-thirds and tool calls down 60%. The feature moves persistent task learning into the agent loop, though the results are still internal.

A new Human-on-the-Bridge paper argued for front-loading expert judgment into reusable evaluation assets, while practitioners also shared double-run and multi-model review setups. The cluster matters because teams tuning agent harnesses need repeatable ways to measure behavior beyond one-off benchmark scores or subjective PR review.

Builders released reusable loop artifacts this week, including a Loop Library Skill, repo templates, and published control-loop definitions for docs sweeps, onboarding checks, and error triage. It matters because teams are turning one-shot prompting into persistent agent runs with explicit stop conditions and shared repo state.

Sakana launched Marlin, a Virtual CSO that runs for up to 8 hours, forms hypotheses, browses sources, and returns slide decks plus reports. It turns Sakana’s long-horizon reasoning work into a shipped deep-research product.

Codex users are having the agent write its own `/goal` and sub-agent goals, with OpenAI-side commentary describing that as a built-in meta-prompting pattern. The workflow turns long autonomous runs into a tighter control loop, but users still review goals first so a bad objective does not burn tokens for hours.

Users are using Fable 5 as a planner and long-run orchestrator while pushing implementation and heavy reasoning to Opus and Codex. The setup keeps Fable on supervision and planning, so teams can track execution through live status pages on larger tasks.

Hyperbrowser shipped a Claude Code harness, InsForge showed a Fable run drop from 5.5M to 2.3M tokens, and Higgsfield published new MCP workflows. These tools add reusable harness, context, and interface layers around Fable for more controlled runs.

Matt Pocock's /teach skill installs with npx skills add mattpocock/skills --skill teach and runs structured strategy lessons inside a Claude agent. Follow-up posts add primary-source reading to the lessons and point to a larger dedicated repo.

Anthropic published Fable-specific guidance for Claude Code and API, emphasizing the /model switch, higher default effort, simpler prompts, and /goal-style verification loops. The Claude Code team says older prompt scaffolds can work against Fable's longer sessions.

Builders shipped OpenProse workflow files, ghzinga PR tabs, cmux terminal controls, datasette-agent-edit primitives, and an agent-optimized CLI fork. These pieces turn prompt strings into reusable files, panes, and testable edit loops for coding agents.

Practitioners shared repeatable setups for multi-hour Claude runs using auto approvals, dynamic workflows, cloud sessions, and critique loops. One large-codebase sweep reported 144 bugs fixed in about four hours with fewer false positives under model critique.

A community workflow broke long-running Codex goals into six required fields, then added an eight-item preflight checklist and helper tools. The structure is meant to reduce runs that drift, stop early, or claim completion without an objective verification step.

Builders added /dynamic orchestration, custom-model routing, and repo runbooks around Codex as users exposed new session lifecycle controls in the app. That makes Codex a better fit for long-running, multi-context coding work.

Three independent Pi builders shipped a goal runner, contract-style subagent acceptance gates, and a new Lovely Dev Tools extension in the same window. That gives Pi users more deterministic long-running loops and cleaner local tool interfaces without starting from an empty harness.

A day after Claude Code introduced Dynamic Workflows, builders shipped ports and clones for Codex, Conductor, and GLM-backed CC Mirror. The rapid ports turn the feature into a reusable orchestration pattern rather than an Anthropic-only runtime.

Firecrawl launched /monitor, a URL watcher that only pings agents when tracked pages actually change and can send results by webhook. Use it for change-only ingestion to cut LLM token spend on monitored pages.

Practitioners published tests-first coding-agent workflows built around red-green TDD, Hurl suites, GitHub label actions, and Codex-based execution checks. The pattern matters because verification remains the main bottleneck once generation is fast, especially in longer multi-file sessions.

Microsoft Research released SkillOpt, which optimizes external skill files instead of fine-tuning model weights and reports best-or-tied results across 52 evaluation cells. The method matters because it improved Codex and Claude Code accuracy without extra inference-time calls.

Practitioners published reusable Codex workflows for project audits, memory-driven skill packaging, mobile delegation, and remote computer use. Try the prompt-and-steps patterns if you want to adapt Codex across repos and devices.

Independent builders published reusable skills infrastructure across coding agents, including Project Think preview support, handoff docs, and an htmx v4 skill pack. That matters because skills are starting to work like portable workflow units instead of one-off prompt snippets inside a single tool.

Independent Codex users published Obsidian memory setups, reusable skill prompts, auto-triage flows, and Cloudflare-backed runners for longer jobs. That matters because Codex is being wrapped into persistent workspaces and operator-defined subagents instead of one-shot chats.