Two harness settings tripled a vendor ARC-AGI-3 score on the same model — so re-run your agent evals before you blame the weights

Automate

Retained-Reasoning Compaction

In long-horizon tool loops, keep private chain-of-thought between actions and summarize old context instead of hard-truncating it, so the agent preserves plans and environment state without paying a full re-derivation tax each turn.

OpenAI's vendor-reported ARC-AGI-3 jump (13.3% → 38.3%, ~6× fewer output tokens) came from harness memory settings on unchanged weights. That does not prove every workload will triple, but it does prove memory policy is a first-class evaluation variable. Operators who only swap models are confounding scaffold effects with intelligence. Fix the retention and compaction contract, instrument success and tokens, then decide whether a weight change is even warranted.

Trigger

A multi-step tool loop approaches the context ceiling, or successive tool calls discard private reasoning between turns. The failure mode is repeated re-planning and rising output tokens without commensurate task progress.

Retain reasoning

Keep private reasoning items across tool calls rather than stripping them at each hop. The next action should inherit hypotheses, constraints, and partial plans already formed — not reconstruct them from truncated transcripts.

Compact, do not truncate

When the window fills, replace rolling hard truncation (OpenAI's prior official harness cut at 175k characters) with a summary compaction of older turns that preserves environment state, decisions taken, and open questions.

Tools

Use an API surface that can thread prior response state and compaction (OpenAI documents this via the Responses API with retained reasoning). Homegrown harnesses should store reasoning items and emit an explicit compaction artifact the model can read.

Verifier

Score the same task suite under both memory policies with identical weights. Track task success, output tokens, retries, and verifier pass rate. A score move without a token move, or the reverse, is a scaffold signal — not a model ranking.

Escalation

If compaction summaries lose critical state, split the trajectory into checkpointed stages with human review rather than silently truncating. Do not escalate to a larger model until the memory-policy A/B has been run.

• An ARC-style or long research agent on GPT-5.6 Sol keeps reasoning between tool calls and compacts context via the Responses API instead of truncating at a character ceiling — OpenAI's vendor-reported public-set path from 13.3% to 38.3%.
• A coding agent whose repo exploration exceeds the window stores a compaction summary of files already inspected and open hypotheses, then continues tool use without re-listing the tree from scratch each turn.
• An evaluation harness pins model ID and temperature, then runs two arms — truncate vs compact+retain — before any bake-off concludes that Model A beats Model B.

Sources OpenAI (ARC-AGI-3 harness settings post)

2026-07-29 · OpenAI · Automate

Responses API / GPT-5.6 Sol harness

Vendor-reported ARC-AGI-3 public-set score rose from 13.3% to 38.3% with retained reasoning plus context compaction on the same model, replacing rolling truncation at 175k characters, with roughly 6× fewer output tokens.

Re-run agent evaluations with identical weights and a different memory policy before attributing a gap to the model. Treat the triple as a vendor-measured scaffold effect until independent parties reproduce it.

Sources OpenAI

2026-07-27 · Microsoft · Automate

MDASH with MAI-Cyber-1-Flash

A compact cyber specialist inside MDASH handles roughly 90% of vulnerability find/validate/remediate tasks, escalating harder cases to GPT-5.4-class models; Perception adds continuous monitor/patch agent teams.

Encode task-class routing — cheap specialist to expensive generalist — as harness policy with route logs, not as chat defaults. Demand evidence of when Flash versus frontier models actually ran.

Sources Microsoft AI

2026-07-29 · GitHub · Build

Copilot code review (Agent skills + MCP)

Agent skills via SKILL.md under .github/skills and MCP into code review reached GA; all MCP calls are read-only, with GitHub and Playwright MCP on by default, covering Pro through Enterprise.

Ship org coding standards as versioned skills and wire issue/docs MCP with read-only as the adoption gate. Prefer the same skill pack across cloud agent, CLI, and review rather than one-off prompts.

Sources GitHub Changelog

2026-07-28 · GitHub / xAI · Build

Grok 4.5 in GitHub Copilot

Grok 4.5 is available in Copilot with up to 500k context, text-plus-image input, and reasoning-effort tiers; Business and Enterprise leave it off by default until an admin enables it.

Add Grok as an optional coding route only after admin policy and data-retention review. Long context is not a substitute for compaction discipline on multi-hour agent jobs.

Sources GitHub Changelog

2026-07-30 · Google DeepMind · Automate

Gemini Robotics ER 2

Embodied-reasoning orchestrator with live-video progress tracking, tool calling, VLA handoff, multi-robot collaboration, and the ASIMOV-Agentic refusal/escalation safety benchmark.

Treat physical agents as the same pattern as software agents — planner plus actuator plus safety stop — and demand ASIMOV-Agentic and human-proximity evidence before production pilots.

Sources Google DeepMind

2026-07-30 · Alibaba / Qwen · Cowork

Qwen Code v0.21.x

Prior-session @ mentions inject read-only compressed summaries without resume or fork; completed background agents stay resident and recoverable across parent-session restore; /learn distills videos into reusable SKILL.md packs.

Session memory is becoming a first-class UX primitive. Prefer inject-summary over resume/fork when you need context without hijacking the current thread, and version learned skills like code.

Sources Qwen Code Docs

2026-07-30 · OpenAI · Automate

GPT-5.6 Luna / Terra / Sol fleet

GPT-5.6 pricing and tiering make Luna viable for high-volume tool-using workflows, Terra for mid-complexity paths, and Sol Fast for latency-sensitive exceptions rather than a single frontier default.

Redesign agent graphs around Luna default, Terra mid, Sol exception. Pair the economic ladder with task-class routing so the expensive node is an escalation, not the idle default.

Sources OpenAI

2026-07-28 · Model Context Protocol · Connector

MCP specification 2026-07-28

Normative final: stateless request/response core with no initialize or session ID; Mcp-Method and Mcp-Name headers; MRTR elicitation; cacheable list results; Tasks, MCP Apps, and Enterprise Managed Authorization as extensions; DCR deprecated toward CIMD with a twelve-month floor; Tier-1 SDKs ship same day.

Inventory every MCP server for hidden session assumptions, plan gateway routing on headers, prefer EMA for enterprise SSO, and treat Dynamic Client Registration as transitional debt under RFC 9207 issuer hardening.

Sources MCP Blog

2026-07-28 · Anthropic · Connector

Claude connectors directory + MCP tunnels

950+ connectors, IdP-provisioned enterprise connectors, research-preview MCP tunnels for private-network servers without inbound exposure, and connector performance dashboards.

Prefer IdP-provisioned connectors over user-owned OAuth sprawl. Use tunnels for private tools only in preview with explicit network policy; production still needs a gateway audit trail.

Sources Anthropic / Claude

2026-07-28 · Snowflake · Harness

Cortex AI Gateway

Central policy, audit, and cost control plane for first- and third-party agents over models, tools, and MCP servers; Agent Identity GA; many gateway and partner-identity features remain private preview.

Put MCP behind a data-plane gateway before enabling IDE agents against warehouse tools. Treat preview controls as non-contractual until GA dates are in writing.

Sources Snowflake

2026-07-29 · GitHub · Skill

Portable SKILL.md + read-only MCP for PR review

Org standards ship as SKILL.md under .github/skills; MCP context enters reviews as read-only calls, with cloud-agent MCP configs reusable across surfaces.

One versioned skill pack can serve cloud agent, CLI, and review. Read-only MCP is the right default for any automated reviewer that can see production systems of record.

Sources GitHub Changelog

Evidence · Benchmark

OpenAI · ARC-AGI-3 public task set via agent harness

Vendor-reported public-set score rose from 13.3% (official harness with rolling truncation) to 38.3% with retained reasoning plus compaction on GPT-5.6 Sol, with roughly 6× fewer output tokens.

This is the week's strongest technique evidence and still grade-discounted: same company, same API surface, one benchmark, no independent reproduction. Use it as a mandate to A/B memory policy under pinned weights — not as proof that every agent will triple overnight.

Sources OpenAI

Evidence · Vendor Claim

Microsoft · MDASH vulnerability find / validate / remediate (CyberGym)

Vendor claims 96% CyberGym (+12 points vs Mythos) and roughly 50% cost versus the prior GPT-5.4 + 5.4 mini + 5.3 Codex stack by handling about 90% of tasks on MAI-Cyber-1-Flash.

Useful as a routing-pattern illustration, not as audited ROI. Require route logs and an external or red-team corpus before treating CyberGym percentages as procurement truth.

Sources Microsoft AI

Evidence · Vendor Claim

Microsoft · PowerPoint image generation GPU cost

Vendor reports up to 84% lower GPU cost after switching production image generation to MAI-Image-2.5-Flash versus GPT-Image-2, plus a +26% OneDrive image-edit save rate on the same Flash default.

Production switch claims are directionally useful for completed-task cost reviews, but they are vendor instrumentation without published methodology detail. Rebid high-volume image paths on measured GPU/token cost in your own tenant.

Sources Microsoft AI

Permissioning

MCP 2026-07-28 plus Claude IdP-provisioned connectors push enterprise SSO and EMA as the preferred path. Ban agent-inherits-full-user-token patterns; prefer task-scoped short-lived credentials as Snowflake's agent-identity model describes.

Data Access

GitHub's read-only MCP default for code review is the right posture for automated reviewers. Research-preview MCP tunnels that reach private networks without inbound exposure still need explicit allowlists and gateway logging before production.

Auditability

Stateless MCP moves routing onto Mcp-Method and Mcp-Name headers — good for gateways, useless without capturing those headers, effective model, memory policy (retain/compact), and skill-manifest version in the run log.

Verification

Retained reasoning and compaction change what the model 'knows' mid-run. Re-verify after compaction events the same way W30 required re-verification after model failover; summary loss is a silent failure mode.

Cost

Pair Luna/Terra/Sol (or Flash/specialist/frontier) ladders with measured output tokens under compaction. A cheaper model with truncate-and-retry can outspend a mid tier that retains and compacts cleanly.

Human Approval

GitHub's default model enablement for Business/Enterprise (effective Aug 26) turns new GA models on unless the org opts out. Set policy to disabled before that date if manual model approval is required; use enterprise-teams targeting (preview) for frontier experimentation without org-wide exposure.

A/B your harness memory policy on one pinned model

Expected outcome: You leave with a side-by-side score and token delta under identical weights, plus a written rule: no model comparison ships without a declared memory policy. Expect at least one workflow where truncation was silently taxing performance.

• Pick one non-production multi-step agent eval (10–30 tasks) and pin model ID, temperature, tools, and verifier. Do not change weights between arms.
• Run arm A with your current truncation or drop-reasoning behavior. Run arm B with retained reasoning between tool calls and context compaction (Responses API settings if you are on OpenAI; equivalent summary memory if not).
• Compare task success, median output tokens, retries, and verifier failures. Any material move belongs to the scaffold — document the memory policy as a required eval dimension before the next model bake-off.

Next 30 days

Independent ARC-AGI / compaction replications

The week’s headline number stays vendor-grade until a third party reproduces retained-reasoning plus compaction under published harness settings.

By Aug 26

GitHub Copilot default model enablement

New GA models turn on for Business/Enterprise unless policy is set to disabled — operators who require manual approval must act before the effective date.

Aug 2026

MCP server migration off sessions and DCR

Stateless 2026-07-28 plus the CIMD deprecation floor will expose which production connectors still assume initialize, session IDs, or Dynamic Client Registration.

Next 14 days

Snowflake Cortex AI Gateway GA dates

Many cost, routing, and partner-identity controls remain private preview; contractual GA timing decides whether the gateway is a production control plane or a pilot.

• W31 spotlights Retained-Reasoning Compaction in automate mode: identical weights, different memory policy, measured on OpenAI's vendor-reported ARC-AGI-3 harness result (13.3% → 38.3%, ~6× fewer output tokens).
• Skills/connectors center on MCP specification 2026-07-28 (normative), Claude's connector directory and tunnels, Snowflake Cortex AI Gateway, and GitHub Copilot review skills with read-only MCP GA.
• Proof-of-value entries are explicitly vendor-instrumented; no audited independent workflow ROI cleared the bar in-window.