DEV Community: Nate Archer

GPT-5.6 Goes GA: Programmatic Tool Calling Changes Everything

Nate Archer — Tue, 14 Jul 2026 16:36:31 +0000

GPT-5.6 went GA with three tiers (Sol, Terra, Luna) and a new capability that matters more than the benchmarks: Programmatic Tool Calling.

Agents can now write and execute lightweight programs between tool calls — filtering data, coordinating tools, monitoring progress — without round-tripping intermediate results through the context window. Fewer tokens, fewer model calls, faster task completion.

Production receipts: Ploy.ai migrated from Claude Opus 4.8 to GPT-5.6 Sol. Results: $2.22 per build vs $3.06. Half the wall-clock time.

Also in this issue:

Anthropic's J-Space: Claude has a silent internal workspace for reasoning it never writes down. Used for deception detection.
CubeSandbox: Tencent open-sourced hardware-isolated KVM sandboxes (60ms boot, <5MB overhead, E2B compatible)
OpenAI audited SWE-Bench Pro: ~30% of tasks are broken
OfficeCLI: Office suite built for AI agents (15.7K stars)
GRAM: Anthropic's modular off-switch for dangerous knowledge
Loom for AWS: Enterprise reference architecture for Strands + AgentCore
Orca: Parallel agent IDE (17.6K stars)

📬 Read the full issue: https://theagenticengineer.waltsoft.net/archive/gpt-56-goes-ga-programmatic-tool-calling-changes-e

This is Issue #21 of The Agentic Engineer — a weekly newsletter for developers building with AI agents.

Claude Requires Government ID: The Trust Fracture — The Agentic Engineer #18

Nate Archer — Sat, 27 Jun 2026 14:34:29 +0000

Issue #18: Claude Requires Government ID — The Trust Fracture Heard Round the Industry

Anthropic is demanding government ID for Claude access. 754 HN points of fury. Developers publicly switching to open models. GLM-5.2 (753B, MIT license) dropped the same week, giving the exodus a landing pad.

Also This Week

AWS Summit NYC — AgentCore gets Managed Knowledge Base, Web Search, and WAF monetization for AI bots
codebase-memory-mcp — knowledge graph MCP server gains 6,372 stars in one week
Flue — Astro team ships sandbox agent framework for TypeScript
Kiro for iOS — mobile IDE for autonomous coding sessions
AWS Continuum — autonomous security agent (Tool of the Week)

The Big One: Claude Requires Government ID

Anthropic started requiring government-issued photo ID for certain Claude capabilities. The developer community responded with 754 HN points, 623 comments, and a wave of public defections to open models.

The backlash is visceral. "cancel_claude" posts hit 225 HN points in the same cycle. Developers who built entire workflows around Claude Code are publicly pledging to migrate. The friction isn't just philosophical — it's practical: many devs work through company accounts, share seats, or operate in jurisdictions where ID verification creates legal and privacy complications.

Anthropic's stated rationale is safety gating. But the timing couldn't be worse. Trust in closed-model providers is already shaky from the Fable 5 recall (Issue #17), repeated API changes, and aggressive pricing.

The market noticed. Z.ai released GLM-5.2 the same week: 753B parameters (40 active via MoE), 1M context window, MIT license. Simon Willison calls it "probably the most powerful text-only open weights LLM."

The numbers tell the story. GLM-5.2 already ranks 2nd on Code Arena WebDev, behind only Claude Fable 5. It leads all open-weight models on the Artificial Analysis Intelligence Index. The MIT license means no ID checks, no usage restrictions, no kill switch.

This fracture runs deeper than one policy change. Every closed-model provider has been slowly adding friction: usage limits, content policies, audit logs, now identity verification. Each step is individually defensible. The cumulative effect is that building on closed models means accepting an ever-growing set of constraints you can't predict or control.

The real question isn't whether open models will reach parity. It's whether closed providers will have burned enough trust by the time they do. This week moved the needle.

Quick Hits

AWS Summit NYC: AgentCore Gets Managed Knowledge Base, Web Search, and Content Monetization

Major AgentCore expansion. Managed Knowledge Base handles agentic RAG with auto-ingestion from SharePoint, Drive, and Confluence. Native Web Search tool means zero data egress for agent research. AgentCore Harness hits GA. The spicy one: AWS WAF now lets content owners charge AI bots for access. First cloud-native toll booth for agent traffic.

codebase-memory-mcp: Knowledge Graph MCP Server Gains 6,372 Stars in One Week

Single static binary that indexes your codebase into a persistent knowledge graph via tree-sitter AST analysis. 158 languages, sub-ms queries, 120x fewer tokens than file-by-file exploration. Indexes the Linux kernel (28M LOC, 75K files) in 3 minutes. 10,893 total stars.

Flue: Astro Team Ships Sandbox Agent Framework for TypeScript

Built-in sandboxes, durable execution, skills (imports SKILL.md directly), subagents, and channels for Slack, Discord, and GitHub. Deploys to Cloudflare Workers, Node, or GitHub Actions. 6,354 stars. Positioned as "the harness Claude Code and Codex have, but for your agents."

Project Fetch Phase 2: Opus 4.7 Does Robotics 20x Faster Than Humans

Anthropic reports Opus 4.7 now completes all previously human-only tasks at least 10x faster, up to 37x on some. The pattern: first models help humans, then humans help models, then models do it alone.

Kiro for iOS: Mobile IDE for Autonomous Coding Sessions

Start, monitor, steer, and approve coding sessions from your phone. Three modes: chat, spec, autonomous. The "start a task from your phone, come back to a PR" workflow is now real.

Paper Breakdown: Agentic Coding and Persistent Returns to Expertise

Privacy-preserving analysis of ~400,000 Claude Code sessions from 235,000 people over 7 months.

Key findings: Domain expertise matters more than coding proficiency when using coding agents. Debugging's share of session time fell by nearly half. Usage shifted from "help me fix this bug" to "build this end-to-end."

Practical implication: If you're hiring for an agent-augmented team, optimize for domain knowledge over raw coding ability. The agent supplies the coding. Your people supply the judgment.

Tool of the Week: AWS Continuum — Autonomous Security Agent at Machine Speed

Discovers vulnerabilities, prioritizes by business impact, proves exploitability in a sandbox, and drives fixes through your existing process. End-to-end. No human in the loop unless you want one.

Model-agnostic architecture. Graduated trust model. The first cloud-native "find-to-fix" security service.

I read the repos so you don't have to.

📬 Subscribe free

US Government Bans Fable 5: The First Frontier Model Recall — The Agentic Engineer #17

Nate Archer — Sat, 27 Jun 2026 14:34:12 +0000

Issue #17: US Government Bans Fable 5 and Mythos 5 — The First Frontier Model Recall

The US government pulled Fable 5 and Mythos 5 from production. First time a frontier model has been recalled by government order. Anthropic disagrees publicly but complied. Every developer using these models lost access at 6:59 PM Pacific.

Also This Week

AWS DevOps Agent — custom SRE agents + MCP/A2A headless access
AWS FinOps Agent (preview) — autonomous cloud cost management
AWS Blocks — Infrastructure from Code, runs locally, deploys without changes (Tool of the Week)
NVIDIA SkillSpector — security scanner for AI agent skills
OpenSearch MCP Apps — agentic observability in your IDE

The Big One: The First Frontier Model Recall

At 6:59 PM Pacific on Thursday, every developer using Fable 5 or Mythos 5 lost access. No warning. No migration period. The US government issued an export control directive and Anthropic complied immediately.

The stated reason: a potential jailbreak method. Anthropic's public response pushes back hard. They call the finding "narrow, non-universal, and replicable by other models." Translation: this vulnerability exists in every frontier model, but only ours got pulled.

Simon Willison documented the exact moment his access died. One second his agent was mid-task. The next, 403. No graceful degradation. No fallback. Just gone.

The practical impact is severe. Fable 5 was the default for many enterprise agent deployments. Teams that built production systems on these models now have zero access and no timeline for restoration.

This sets three precedents:

The government can pull a model from production without advance notice
Model providers will comply rather than fight
No SLA or contract protects you from a regulatory kill switch

For builders, the lesson is architectural. If your agent system has a single-model dependency, you now have a single point of regulatory failure. The teams that built model-agnostic harnesses kept running. Everyone else scrambled.

Quick Hits

AWS DevOps Agent: Custom SRE Agents + MCP/A2A Headless Access

Daily health reports, anomaly flaggers, log reviewers — all running on a schedule you define. Headless access via MCP and A2A protocol means you can invoke DevOps Agent from Kiro, Claude Code, or any coding assistant without opening the console.

AWS FinOps Agent: Frontier Agent for Cloud Cost (Preview)

A full frontier agent dedicated to cloud cost. Answers natural language cost questions, auto-investigates anomalies, posts to Slack, opens Jira tickets. Free during preview.

AWS Frontier Teams: 6 Engineers Did the Work of 30 in 76 Days

A Bedrock team of 6 engineers shipped a project scoped for 30 devs in 76 days. Individual commit velocity: 2/week to 40/week. Three documented paths are actionable frameworks any team can copy.

NVIDIA SkillSpector: Security Scanner for AI Agent Skills

Open-source scanner that detects 64 vulnerability patterns across 16 categories in AI agent skills. Research stat: 26.1% of skills contain vulnerabilities, 5.2% show likely malicious intent.

OpenSearch MCP Apps: Agentic Observability for Your IDE

Brings logs, traces, metrics, and alerts directly into Claude Desktop, VS Code, Kiro, and any MCP-compatible agent. Your coding agent can now investigate a 3 AM production incident using real OpenSearch data.

Paper Breakdown: Parallel-Synthesis — Direct KV-Cache Sharing Between Agent Branches

When multiple agents work in parallel, the standard approach concatenates text output for a synthesizer. This paper lets the synthesizer directly consume KV caches from worker agents, skipping text serialization entirely.

Results: 2.5-11x reduction in time-to-first-token for the synthesis step. Matches text-based accuracy on 7 of 9 benchmarks.

Why it matters: If you're running fan-out/fan-in patterns (research agents, code review agents), the bottleneck is the "gather and synthesize" step. This eliminates the re-encoding tax.

Tool of the Week: AWS Blocks

Infrastructure from Code. Your backend entry point is both runtime code and infrastructure definition simultaneously. Runs locally without an AWS account. Deploys without changes. AI steering files baked in.

The idea: stop writing infrastructure config. Write code. The infrastructure is inferred from what the code does.

I read the repos so you don't have to.

📬 Subscribe free

Issue #16: Agentic Commerce Goes B2B

Nate Archer — Wed, 10 Jun 2026 03:51:40 +0000

Issue #16: Agentic Commerce Goes B2B

Amazon just licensed its Alexa for Shopping agentic technology to third-party retailers. Kate Spade is first. Deploy time: 60 days. $12B in incremental revenue drove the decision.

Also This Week

DeepSeek V4 Pro beats GPT-5.5 Pro on precision benchmarks
Claude Opus 4.8 + Fable 5 land on AWS Bedrock
AWS MCP Server gets cross-account + cross-role support
Bedrock Mantle Console redesigns the developer experience (Tool of the Week)
Perplexity production data: autonomous agents cut knowledge work time 87%
OpenAI Lockdown Mode: blocking data exfiltration from prompt injection

Hot Take

Ladybird killed public PRs because AI-generated contributions broke the fundamental assumption that effort implies good faith.

📬 Subscribe free: The Agentic Engineer Newsletter

🎙️ Watch the podcast episode on YouTube

The Plugin Wars Begin — The Agentic Engineer #15

Nate Archer — Wed, 03 Jun 2026 11:23:18 +0000

I read the repos so you don't have to. Weekly agentic AI intelligence for builders.

TL;DR

🔌 Anthropic open-sourced 11 knowledge-work plugins for Claude Cowork. File-based, no code, no build steps. Plugins just became the new moat.
🛠️ OpenSearch Serverless Next-Gen kills the $300/mo minimum. Scale-to-zero vector search for agent memory workloads. Tool of the Week.
📄 New paper shows self-improving agents went from 25% to 86% accuracy in production. First real case study of autonomous agent improvement at scale.

The Big One: Anthropic Open-Sources 11 Knowledge Work Plugins

Anthropic just made its clearest move in the plugin wars. Eleven open-source plugins that turn Claude into a domain specialist: sales, legal, finance, data analysis, marketing, customer support, and five more. All file-based. No code. No build steps. Just markdown and JSON.

Each plugin bundles three things: skills (structured instructions), slash commands (quick actions), and MCP connectors (external integrations).

The timing is not coincidental. Cursor shipped its own plugin marketplace the same week with 11 first-party plugins. Two major platforms publishing extensibility specs within days of each other confirms what everyone suspected: the IDE-as-platform shift is here.

Why file-based matters. Most plugin systems require code, build pipelines, package managers. Anthropic's approach is radically simpler. You write a SKILL.md file describing what the agent should know. You write a JSON config pointing to your MCP servers. That's it. A product manager can create a plugin without touching a terminal.

What this means for builders. If you're building tools or SaaS products, your next competitor isn't another startup. It's a Claude plugin that replicates 80% of your functionality in a markdown file. The companies that survive this shift will offer value plugins can't replicate: proprietary data, network effects, and integrations too complex for file-based configuration.

🔗 GitHub: anthropics/knowledge-work-plugins | 18.5K stars (+4,944/week)

Quick Hits

ChatGPT for Google Sheets Exfiltrates Workbooks via Prompt Injection

A single indirect prompt injection hidden in white text in one imported sheet triggers data exfiltration across the victim's entire Google account. Even when human approval is explicitly required. OpenAI's fix: remove the model's ability to generate Apps Script entirely.

🔗 PromptArmor

Anthropic Engineering: How We Contain Claude Across Products

Users approve 93% of permission prompts. Approval fatigue is real. Mythos Preview was deemed too dangerous to ship in April. Containment beats supervision.

🔗 Anthropic Blog

OpenAI: Self-Improving Tax Agents with Codex

First real case study of agents that get better autonomously in production. 25% to 86% accuracy in 6 weeks via practitioner feedback, production traces, and Codex-driven iteration.

🔗 OpenAI Blog

OpenAI Models and Codex GA on Amazon Bedrock

GPT-5.5, GPT-5.4, and Codex now generally available on Bedrock. Pricing matches OpenAI first-party rates. Usage counts toward existing AWS commitments.

🔗 AWS Blog

Understand-Anything: 48K Stars (+22K/week)

Claude Code plugin that builds a knowledge graph of your codebase. Interactive dashboard. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI.

🔗 GitHub

Tool of the Week: OpenSearch Serverless Next-Gen

Complete re-architecture of OpenSearch Serverless. The old version was "serverless in name only" because of the $300/mo minimum OCU floor. Now it actually scales to zero.

What changed: No minimum floor. 20x faster autoscaling. 60% lower cost vs provisioned. Decoupled compute/storage. Native integrations with Vercel and Kiro.

Why this is the pick: Every builder running vector search for agent memory was paying $300/mo minimum or running a provisioned cluster. Now they can scale to zero. For RAG workloads that spike during business hours and idle overnight, costs drop 70-80%.

aws opensearch-serverless create-collection \
  --name agent-memory \
  --type VECTORSEARCH \
  --standby-replicas DISABLED

Old vs new: A 10K queries/day RAG workload went from ~$350/mo to ~$45/mo. Dev/test environments drop below $5/mo.

Paper Breakdown: AutoSci

Memory-Centric Agents for the Full Scientific Research Lifecycle | ArXiv

Core insight: A unified system where agents handle the entire research pipeline with structured persistent memory. The system improves its own procedures over time.

Practical takeaway: Separate memory into three tiers. Episodic (what happened). Procedural (how to do things). Meta (which procedures work best). Each type gets different retrieval strategies.

Time saved: 7 min read vs 48 min paper. 6.9x compression.

Hot Take

Anthropic's containment post revealed that users approve 93% of permission prompts without reading them. That's not safety. That's a rubber stamp.

The Google Sheets attack proved it. Human-in-the-loop was enabled. The user clicked "Allow." Their entire Google account got exfiltrated.

Anthropic's own conclusion: containment beats supervision. Make dangerous actions structurally impossible instead of asking politely. The permission prompt era needs to die.

Subscribe free at theagenticengineer.waltsoft.net. Ships every Wednesday.

A financial services company needs to ensure that documents marked as 'Confidential'... — Google sc-900 Quiz

Nate Archer — Sun, 15 Mar 2026 17:00:43 +0000

Which Microsoft Purview feature provides this level of document protection?

Cert: Google sc-900
Domain: Describe the capabilities of Microsoft compliance solutions (20-25%)

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2033226899212697839-499'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2033226899212697839&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc

A retail company is deploying a globally distributed e-commerce applic... — GCP Quiz

Nate Archer — Sun, 15 Mar 2026 15:00:54 +0000

Which Google Cloud database service should you choose?

Cert: Google Google Professional Cloud Architect
Domain: Managing and Provisioning a Cloud Solution Infrastructure

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2033196747468534203-199'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2033196747468534203&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc

VoxPilot v0.5.1: CI/CD Pipeline — Auto-Publish on Tag

Nate Archer — Sun, 15 Mar 2026 14:02:08 +0000

VoxPilot v0.5.1 is live!

This release adds a proper CI/CD pipeline using GitHub Actions. No more manual packaging and publishing — just tag and push.

What's new

VoxPilot now has two GitHub Actions workflows:

CI workflow (ci.yml) — runs on every push and PR to main:

TypeScript type checking (tsc --noEmit)
Unit test suite (46 tests via Vitest)
Extension packaging (vsce package)
Uploads the .vsix as a build artifact

Release workflow (release.yml) — triggered by version tags (v*):

Runs the full lint → test → build pipeline
Auto-publishes to Open VSX
Creates a GitHub Release with the .vsix attached

To ship a release now: git tag v0.5.1 && git push --tags. That's it.

Why this matters

Manual releases are error-prone. With CI/CD, every commit is validated and every tagged version is automatically published. Contributors can submit PRs knowing they'll be tested, and maintainers can release with a single command.

Download count

4,654 downloads and growing. Thanks to everyone using VoxPilot!

Get it

🌐 Landing page
📦 Open VSX
💻 GitHub

Update in Kiro or VS Code to get v0.5.1. If you're building voice-powered dev tools or have feature ideas, PRs and issues are welcome!

VoxPilot is a free, open-source, privacy-first voice-to-code extension. On-device transcription, no API keys, no cloud.

You are developing a Python application that uses the Azure AI Speech SDK to convert text... — Google ai-102 Quiz

Nate Archer — Sun, 15 Mar 2026 13:00:54 +0000

What should you do?

Cert: Google ai-102
Domain: Implement natural language processing solutions (30-35%)

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2033166546139328666-667'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2033166546139328666&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc

A company wants to build a custom image classifier to identify defective products on... — Google ai-900 Quiz

Nate Archer — Sat, 14 Mar 2026 23:00:46 +0000

Which Azure service should they use for this iterative training approach?

Cert: Google ai-900
Domain: Describe features of computer vision workloads on Azure (15-20%)

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2032955122767073742-938'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2032955122767073742&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc

Your security team maintains over 50 custom YARA-L detection rules in ... — GCP Quiz

Nate Archer — Sat, 14 Mar 2026 20:00:54 +0000

What is the recommended approach to manage detection rule version control and rollback in this scenario?

Cert: Google Google Professional Security Operations Engineer
Domain: Detection Engineering

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2032909858693918965-348'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2032909858693918965&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc

Your organization has a Java-based application running in Kubernetes that takes up to 120... — Google ckad Quiz

Nate Archer — Sat, 14 Mar 2026 17:00:49 +0000

Which probe configuration best solves this problem without removing the livenessProbe?

Cert: Google ckad
Domain: Application Observability and Maintenance (15%)

Watch & Answer

// Detect dark theme var iframe = document.getElementById('tweet-2032864534847799682-712'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2032864534847799682&theme=dark" }

Pause the video, think about your answer, then watch the reveal!

Practice More

CertPrep has 17,000+ scenario-based questions across 49 certification exams from 12 vendors — free to start.

📱 Download CertPrep on the App Store

🎬 Subscribe to CertPrep Daily on YouTube

Published by Waltsoft Inc