ブログ · 現場のAI実装ノート

Production AI Agent Architecture: 5 Reusable Patterns from DoorDash & Shippy for Shipping Agents to Production

Most enterprise agents die between the demo and production. Two fresh first-party cases — DoorDash's "Ask DoorDash" assistant and the Allen Institute's Shippy — show the gap isn't which model you picked. It's the five engineering things around the model: deterministic tools, an MCP layer that decouples orchestration from business logic, layered memory, evaluating the agent (not the model), and an FDE-style split between domain and platform teams. Here's the reference architecture, with the numbers.

2026年8月2日9 分で読了

From Model Race to Value Loop: How AI FDE Delivery Drives Enterprise AI ROI in 2026

In 2026 a growing argument in enterprise AI is that the contest is no longer about who has the strongest model, but who can close the loop from tokens to business value. That's where the AI Forward Deployed Engineer (FDE) comes in — an on-site delivery model that vendors like Runjian price on the value it actually creates. Here's what an AI FDE is, how it compares with traditional outsourcing, and what verifiable, auditable delivery looks like in practice.

2026年8月2日10 分で読了

The Hidden Costs of Enterprise AI Agents: Why Agents Cost 10–100× More Than Chatbots—and How to Govern It

An AI agent is not a more expensive chatbot—it is a different cost curve. Because every reasoning step re-sends its context, agents burn 10–100× more tokens than chatbots, and the token bill is only the visible layer. Here's the mechanism, the three hidden costs enterprises miss, a self-build vs. platform vs. FDE comparison, and a four-lever governance playbook that cuts spend 50–70% within two weeks.

2026年8月2日9 分で読了

From Model Training to the Delivery Closed-Loop: Enterprise AI Implementation and the AI FDE Model (WAIC 2026)

WAIC 2026 made one thing clear: the bar for enterprise AI has moved from "can the model answer?" to "can it finish the job and be paid on the result?" This is a practical guide to the delivery closed-loop, the training-vs-inference budget question, and the AI FDE (Forward Deployed Engineer) model that ties tokens to business value.

2026年8月2日9 分で読了

AI-Native Methodology

Enterprise AI Agent Security: Securing the New Attack Surface & Non-Human Identity (2026)

AI agent security is not software patching. Autonomous decision chains, a non-human identity explosion, and supply-chain write access make agents a new attack surface. Here is how enterprises secure them in production—with least privilege, NHI governance, bounded exploration, and evidence governance—grounded in the July 2026 Hugging Face intrusion, Cyera's $1B Oasis Security deal, and hard benchmark data.

2026年7月30日12 分で読了

Build, Buy a Platform, or Hire an FDE? How to Choose Your Enterprise AI Adoption Path in 2026

Small businesses and one-person companies adopting AI in 2026 have three real paths: build in-house, buy a general agent platform (Nano Work / ChatGPT for small business), or hire a forward-deployed engineer (FDE). A cost / speed / control / scale comparison table shows exactly which one fits — and where each hits a wall.

2026年7月30日8 分で読了

Why Enterprise AI Agents Fail in Production: 6 Root Causes and the FDE Fix (2026)

Most enterprise AI agents never reach production not because the model isn't smart enough, but because strategy, data, and safety break down first. Gartner expects 40%+ of agentic AI projects cancelled by 2027 and, per Gartner (via FutureAGI), 85% of AI projects never reach production. Here are the 6 root causes—and how forward-deployed engineering fixes each.

AI Agent Security in Production: A Non-Human Identity Governance Playbook

Every AI agent you deploy is a new hire who just got the keys to your systems — but one you never onboarded, never gave an MFA prompt, and can spin up many copies of near-instantly. As agents proliferate, their "non-human identities" have become the enterprise's newest attack surface. Here is an FDE-grade playbook for governing agent permissions in production: give every agent its own identity, grant least privilege, authorize each tool call, keep a human in the loop for high-risk actions, and log the whole decision chain.

AI-Native Methodology

Enterprise AI Agent Security: Governing Non-Human Identities (NHI) Before Your Agents Overstep

Once you run AI agents in production, every agent becomes a non-human identity (NHI) you have to govern — and the security boundary shifts from "a person logging in" to "what identity an agent holds, what it's allowed to do, and whether its behavior is monitored and auditable." Here's the four-part governance framework, grounded in this month's $1B NHI acquisition, a real 17,600-action agent intrusion, and how the big vendors are already fighting agents with agents.

OpenAI Presence: Why Model Vendors Are Switching to Forward Deployed Engineers for AI Agents

OpenAI is delivering its new Presence product through Forward Deployed Engineers rather than a self-serve API — a signal that enterprise AI-agent delivery is moving from "sell an API" to "embed on-site." We unpack the shift and give you a four-factor framework for choosing between a model vendor's FDE team and an independent implementation partner.

Enterprise Agent Platform: The 6 Deployment Requirements That Decide Success

When an AI agent goes wrong, it doesn't say the wrong thing—it *does* the wrong thing. That single distinction is why enterprise agent platforms succeed or fail on engineering, not on model IQ. Here are the six deployment requirements—cost, setup, permissions, reliability, security, and continuous evolution—and how 360, GitLab 19.2, and Meta each answer them.

Securing Enterprise AI Agents: Identity, Permissions & Isolation When Agents Act on Their Own (2026)

AI agents that can act on their own are simultaneously a new attack surface and a machine-speed attacker. This guide breaks down what actually changes when you put autonomous agents into production — non-human identity, permission scope, isolation — and gives you an FDE-grade governance checklist, grounded in a $1B security acquisition and a documented 17,600-action agent intrusion.

OpenAI Presence & the Rise of the FDE: Who Should Deploy Your Enterprise AI Agents?

OpenAI now sends its own Forward Deployed Engineers on-site to stand up enterprise agents through a product called Presence. That quietly turns "build vs. buy" into a three-way decision—AI lab, in-house team, or independent partner—and the right answer depends on control, lock-in, and whether you've built the evaluation and governance to run agents in production at all.

2026年7月30日8 分で読了

Why OpenAI's Presence Ships Only via Forward Deployed Engineers — The FDE Signal for Enterprise AI

OpenAI won't let enterprises buy its new Presence agent off the shelf — it deploys only through Forward Deployed Engineers and systems integrators. That decision is a signal: as models commoditize, the hard part (and the margin) moves to deployment. Here's what the FDE shift means, and how any company can reuse the playbook.

2026年7月30日7 分で読了

What Is an AI FDE? From Selling Models and Tokens to On-Site Delivery and Value-Based Pricing

A comparison-table explainer of how the AI FDE model differs from outsourcing, license sales, and consulting — grounded in named 2026 enterprise data.

2026年7月27日8 分で読了

Enterprise AI Adoption in 2026: 88% Are Using It — So Why Do Fewer Than 10% Actually Scale?

In 2026, enterprise AI is wide adoption but narrow production: 88% of organizations use AI, yet fewer than 10% scale an agent in any single function and only 39% see ROI — while a separate 88% of agent pilots never reach production. A data-driven reality check that keeps the two 88%s apart, plus the path into the 12% that actually ship.

2026年7月27日9 分で読了

x402 Agent Payments Explained: What Enterprises Must Build When AI Agents Pay for Themselves

In one week, AWS and Cloudflare both put the x402 agent-payment protocol into their edge networks — AI agents can now settle in USDC with no account and no API key. The infrastructure layer is now largely built for you. The part still unowned — invoices, VAT, and compliance for machine-to-machine payments — is the part your enterprise has to build.

2026年7月14日8 分で読了

Why Production AI Agents Fail: Reliability, Observability, and the Last Line of Defense

Most production AI agent failures are misdiagnosed. When an agent seems to "get dumber," it's usually configuration drift — not a weaker model — and when one deletes a database, the root cause is over-broad permissions plus no immutable backup. Using Anthropic's own 6,852-session analysis (as reported by 36Kr), an InfoQ case study on Snowflake's WORM last line of defense, and real migration data from ploy.ai, we lay out a four-layer reliability model for running agents in production.

2026年7月13日8 分で読了

Rent an LLM API or Build Your Own AI Agent? A 2026 Enterprise Build-vs-Buy Framework

Renting a frontier API and owning your own agents aren't a binary — they're points on a curve that moves with scale. Here's an answer-first 2026 framework for when to rent, when to build, where the cost inflection point actually sits, and why the companies that cross the adoption threshold hire more, not fewer, people.

2026年7月12日9 分で読了

The Last Mile to Production AI Agents: Observability, Runtime Audit & Evaluation (2026)

Getting an AI agent to run is not getting it to production. The blocker is no longer "can it complete the task" but "can we prove it behaved." Here are the three guardrails that close the last mile—runtime observability, a tamper-evident audit trail, and evaluation contracts—grounded in this month's evidence, from hash-chained audit logs to a 270-run contract pass rate.

2026年7月11日9 分で読了

Why AI Agents Fail in Production (and How Enterprises Ship Them Anyway)

AI agents that dazzle in a demo often break the moment they touch real work. The gap is rarely the model — it's the missing production engineering around it. This guide breaks down the four real reasons enterprise agents fail and the FDE checklist for shipping one anyway.

2026年7月10日7 分で読了