# Baize AI

> Baize AI is an AI transformation consultancy. It helps companies put AI into operations, engineering, and delivery workflows. Work is remote-first.

Baize AI offers four services:

01 · AI transformation consulting: interview each team, rank the workflows to change first, and agree how results will be measured.
02 · Team AI training: role-based sessions; each team leaves with reusable skills for its own tasks.
03 · Forward deployed engineering: remote-first engineers work inside your team and connect agents to CRM, ERP, and HR systems.
04 · Internal tools and product support: custom internal tools, maintained after launch.

Contact: contacts@fxai.ai

## Key pages

- [Home](https://fxai.ai/en/): Who Baize AI is, the four services, and how to book a first call
- [Services](https://fxai.ai/en/services/): Consulting, training, forward deployed engineering, and tools: scope and deliverables
- [Method](https://fxai.ai/en/method/): How an engagement runs, from diagnosis to rollout
- [About](https://fxai.ai/en/about/): The team and the Formal eXplainable AI working standard
- [Notes](https://fxai.ai/en/notes/): Notes on putting AI to work in real teams, filterable by topic
- [Contact](https://fxai.ai/en/contact/): Book a call by email or the contact form

## Notes

- [Can a Computer-Use Agent Safely Operate Your Legacy ERP?](https://fxai.ai/en/notes/agent-legacy-desktop-automation/): Computer-use agents can click through legacy ERP clients with no API. Let them query and save drafts; keep posting and master data edits behind human approval.
- [How to Run a Private Codex Gateway for Your Company on Official API Keys](https://fxai.ai/en/notes/ai-coding-compute-sourcing/): Run a private Codex gateway on official API keys: deploy LiteLLM with Docker Compose behind HTTPS, issue one key per employee with budgets, and log usage.
- [AI Data Queries Return Two Sales Figures: Which Metric Definition Do You Trust?](https://fxai.ai/en/notes/ai-data-query-metric-definitions/): When natural-language data queries return two different sales figures, the fix is a governed metric dictionary the model must check before it writes SQL.
- [AI Resume Screening: Should a Human Review the Candidates the Agent Rejects?](https://fxai.ai/en/notes/ai-resume-screening-human-review/): A recruiting agent speeds up resume screening, but rejected candidates should not be archived unseen. Set up borderline spot checks, appeals, and audit logs.
- [AI Safety Inspection Flags a Hazard: Who Signs Off on the Fix?](https://fxai.ai/en/notes/ai-safety-inspection-hazard-fix/): Factories and construction sites use AI safety inspection to capture on-site hazards. The algorithm captures risk images. A safety officer verifies, grades, and assigns each case. The responsible team fixes it by the deadline, and the officer closes it after an on-site recheck.
- [Who Owns a Missed Defect in AI Visual Inspection, and How Much Manual Checking Should Stay?](https://fxai.ai/en/notes/ai-visual-inspection-spot-check/): When AI visual inspection lets a defect through, saved images and logs decide who is liable. How to split model and inspector duties and tune manual spot checks.
- [When an AI Agent Drafts Your Bid, Who Guards Against Disqualification?](https://fxai.ai/en/notes/bid-agent-tender-response/): Companies use bid agents to help prepare tender responses. One mistake on a mandatory clause disqualifies the bid. The agent lists disqualification risks and drafts responses. Commercial and technical staff check each item against the tender documents. People approve the price, the original credentials, and the signed submission.
- [How to Record Lead Source When AI Sorts WhatsApp and Email Inquiries](https://fxai.ai/en/notes/export-inquiry-whatsapp-email-attribution/): When AI triages B2B inquiries from WhatsApp, email, and web forms, the lead source field needs fixed rules. Five steps keep channel attribution accurate.
- [AI Equipment Fault Diagnosis: Getting Veteran Technicians' Know-How into Work Orders](https://fxai.ai/en/notes/maintenance-work-order-ai-diagnosis/): Manufacturers can hand first-pass fault diagnosis to AI. The lasting gain is logging veteran technicians' verified fixes in work orders as searchable knowledge.
- [How Detailed Should a Procurement Agent's Supplier Shortlist Reasons Be?](https://fxai.ai/en/notes/procurement-agent-supplier-shortlist/): A procurement agent's supplier shortlist needs reasons buyers can verify and auditors can trace, tied to certificate dates, report numbers, and delivery records.
- [AI Flags an SPC Anomaly: Who Decides Whether to Stop the Line?](https://fxai.ai/en/notes/spc-anomaly-ai-line-stop/): Plants hand process data to AI for SPC anomaly detection. When parameters drift, the stop-the-line decision needs a clear split between system and people. The system reads data continuously and suggests causes. Shift leads and quality engineers respond by preset tiers, and people own the stop decision and the restart check.
- [Agent Authentication: Whose Permissions Apply When Calling Systems for Employees?](https://fxai.ai/en/notes/agent-acting-on-behalf-of-user/): Agent authentication pairs employee and agent IDs per request and intersects gateway and employee rights to stop shared-account data overreach and audit bypass.
- [How to Write a B2B Landing Page for AI Agent Buyers](https://fxai.ai/en/notes/agent-buyer-b2b-landing-page/): AI agents now read vendor pages, check specs, compare prices, and send inquiries for B2B buyers. Humans still sign off. What a landing page needs for each reader.
- [Agent Evaluation: How Many Runs per Task Are Enough?](https://fxai.ai/en/notes/agent-eval-suite-pass-rate-variance/): Agent evaluation repeats fixed real business tasks, reports pass rate, cost per run, latency and variation, and lists failures. One run cannot justify launch.
- [AI Agent Security Governance Must Come Before Scaling](https://fxai.ai/en/notes/agent-governance-before-scale/): Deloitte: 23% of firms use AI agents moderately or more; 74% expected in two years. Only 21% rate governance mature. Four checks without longer approval chains.
- [Who Owns an AI Agent When the Employee Who Built It Leaves?](https://fxai.ai/en/notes/agent-owner-when-employee-leaves/): Agents on shared workflows belong to the process owner. When the builder leaves, swap credentials, check triggers, clear private memory; if unowned, freeze it.
- [Your AI Agent Pilot Works. What Does It Need Before Production?](https://fxai.ai/en/notes/agent-pilot-to-production-checklist/): Agent pilots run on clean data with engineers watching. Production brings messy data, cross-system writes, and real consequences. Five checks before you scale.
- [AI Agent Permissions: Reading Email Is OK. Does Sending Need Approval?](https://fxai.ai/en/notes/agent-read-mail-vs-calendar-send/): AI Agent permissions: ease approval for mail search and summaries; approve sends and calendar edits by default. Fill in a permission table before Always allow.
- [AI Agent Rollback: How to Undo a Bad Action and Who Is Accountable](https://fxai.ai/en/notes/agent-rollback-and-accountability/): An AI agent took a wrong action. Stop it and revoke credentials in hour one, roll back by action type, replay the action log, and assign accountability by role.
- [Agent Test Environment: Use Synthetic Data Before Connecting a CRM?](https://fxai.ai/en/notes/agent-sandbox-test-vs-prod-data/): Test agents in mock companies before CRM access. Tools like Era generate them with mock APIs and MCP endpoints. Retest with real data in staging before launch.
- [Agent Management Starts With an Inventory: Who Runs Employee-Built Agents?](https://fxai.ai/en/notes/agent-sprawl-inventory/): Log agent purpose, builder, owner, systems, credentials, triggers, last run. Tier by data sensitivity and action impact. Disable unowned agents by a deadline.
- [Agent Safety Guardrails: When to Stop an Unattended Run](https://fxai.ai/en/notes/agent-stop-conditions-decision-log/): Agent safety guardrails: completion check; caps on steps, tokens, time, repeat calls and tool error streaks. Limit exits are incomplete; log why and open items.
- [Multi-Agent Security: Can Agents Follow Forwarded Instructions Directly?](https://fxai.ai/en/notes/agent-to-agent-trust-boundary/): Agent handoffs may carry malicious external content. Verify originators, limit actions, block internal addresses, and get human approval for high-risk writes.
- [Agent Tool Call Auto-Approval: Which Calls to Allow or Block](https://fxai.ai/en/notes/agent-tool-call-auto-approval-rules/): Agent tool approval: code boundary checks, rule auto-approval, human review. Use arguments; ignore model reasons. Strict write checks. Log approvals and blocks.
- [Agent Permission Management: How to Set Tool Access Levels and What to Grant First](https://fxai.ai/en/notes/agent-tool-permission-ladder/): Agent access by impact: reads, internal drafts, capped writes; humans approve external actions, payments and irreversible operations; upgrade and revocation rules per tier.
- [How to Measure AI Agent Value After Rollout: 4 Metrics Beyond Usage](https://fxai.ai/en/notes/agent-value-work-finished-metrics/): Call counts and active users don't show what an AI agent delivers. Track cycle time, rework rate, tool-switching cost, and agent overlap against team baselines.
- [AI Customer Service Auto-Close: Why Tickets Still Need Human Review](https://fxai.ai/en/notes/ai-cs-auto-close-vs-human-ticket-review/): An AI bot marking a chat resolved does not clear the ticket. How to split auto-close from human ticket review, as China's GB/T 47746-2026 standard does.
- [AI Customer Service Compliance: Who Owns What the Bot Promises?](https://fxai.ai/en/notes/ai-cs-can-reply-vs-who-owns-liability/): China's GB/T 47746-2026 took effect Sept 1, 2026. AI agents may answer order questions; refunds, discounts, and compensation need human or formal sign-off.
- [Is OpenAI Retracing Dropbox's Path? AI Monetization Starts With Who Pays](https://fxai.ai/en/notes/ai-monetization-dropbox-lesson/): Pricing a consumer AI app starts with who pays. Ben Thompson says OpenAI is retracing Dropbox's path, building ad tracking and enterprise products at once.
- [AI Native vs. AI-Enhanced: Does Your System Still Work Without the AI?](https://fxai.ai/en/notes/ai-native-vs-ai-enhanced/): Remove the AI from an AI-enhanced system and it still runs, only slower. Remove it from an AI Native system and nothing is left. One test to tell them apart.
- [How to Set an Enterprise AI Strategy: Amazon, Apple, and Microsoft's Three Bets](https://fxai.ai/en/notes/ai-strategy-first-customer/): Start an AI strategy by picking your role. Amazon is its own first customer, Apple owns the entry point and buys models, Microsoft builds the middle layer.
- [Can an Annual IT Budget Handle Usage-Based AI Pricing?](https://fxai.ai/en/notes/ai-usage-pricing-budget/): Microsoft 365 E7 is $99 per user per month, with agent overage billed in Copilot Credits. Usage pricing needs monthly decisions; most budgets are set yearly.
- [Choosing Enterprise AI Use Cases: Start With Work You Can Verify on the Spot](https://fxai.ai/en/notes/ai-verifiable-work-first/): Code and math can be checked instantly; long-cycle work cannot. Rank enterprise AI use cases by feedback loop length. Digital ads already run a closed loop.
- [AI Video Ads: Producing More vs. Passing Brand Review](https://fxai.ai/en/notes/ai-video-ads-vs-brand-review/): Models can make dozens of AI video ads a day; fewer pass brand and platform review. China's labeling rules, Google's AI ad labels, and a pre-scale review gate.
- [AI Outbound Calling vs. Human Telesales: Which Steps of a Lead List Go to Each?](https://fxai.ai/en/notes/ai-voice-outbound-vs-human-telesales/): Split one lead list by task when buying AI outbound calling. AI handles bulk reach, screening, and standard Q&A. Humans handle objections, pricing, and closing.
- [What Should a Legal Ledger Track for AI-Generated Ad Creatives?](https://fxai.ai/en/notes/aigc-ad-assets-vs-legal-ledger/): China's AI content labeling rules took effect 2025-09-01. An ad legal ledger needs source, generation process, and license fields, plus who logs each one.
- [Company-Wide AI Training Is Done. Why Haven't Workflows Changed?](https://fxai.ai/en/notes/all-hands-ai-training-workflow-unchanged/): McKinsey's 2025 survey: 88% of organizations use AI regularly, yet most haven't embedded it in workflows. After AI training, redesign one real workflow first.
- [How to Write Answer-First Content That Answer Engines Can Quote](https://fxai.ai/en/notes/answer-first-for-answer-engines/): Answer engines often quote the first paragraph under a question-style H2. Write the full answer in the first sentence, then the reasons and exceptions after it.
- [AP Invoice Agents That Also Pay: Which Controls Can't Be Skipped?](https://fxai.ai/en/notes/ap-invoice-agent-payment-control/): An AP agent that runs from invoice matching to payment needs match records, call-back checks on bank changes, duplicate checks, limits, batch review, and recon.
- [How to Write B2B Comparison Pages That AI Assistants Cite](https://fxai.ai/en/notes/b2b-comparison-page-ai-citation/): Siege Media data on 116 B2B sites: X vs Y pages correlate most with AI referrals. Lead with a verdict, a decision-axis table, fair trade-offs, stable names.
- [Should You Paste a Contract into ChatGPT? Redact It First](https://fxai.ai/en/notes/chatgpt-contract-paste-vs-redact/): CISA's acting director uploaded a For Official Use Only contract to public ChatGPT. Consumer plans may train on inputs. Redact names, amounts, and seals first.
- [How to Get ChatGPT and Perplexity to Cite Your Website](https://fxai.ai/en/notes/chatgpt-perplexity-cite-your-site/): To get cited by ChatGPT and Perplexity, allow OAI-SearchBot, PerplexityBot, and Perplexity-User, then give key pages a short answer that crawlers can extract.
- [What Does It Take to Make Codex Part of Your Team's Dev Workflow?](https://fxai.ai/en/notes/codex-into-real-workflow/): Moving Codex from one engineer's side tool into a team's development process means settling four issues - task scope, verification, ownership, and lessons.
- [Codex Subagent Models: Keep Judgment on the Main Thread, Send Repeat Work to Cheaper Tiers](https://fxai.ai/en/notes/codex-subagent-model-tiers/): How to assign GPT-6.1 Sol, GPT-6 Astra, and GPT-6 Luna to Codex subagents by role, which setting wins, and which reasoning levels each model supports.
- [Why Coding Agent Spend Doesn't Reconcile With Company Books](https://fxai.ai/en/notes/coding-agent-spend-audit-on-gateway/): Codex and Claude Code spend on personal keys won't match finance. Who reconciles, which log fields to keep, and why only gateway-issued credentials work.
- [Which CRM Outbound Agent Emails Need Human Approval Before Sending?](https://fxai.ai/en/notes/crm-outbound-agent-send-approval/): First touches, pricing, key accounts, and unclear context go to a human. Routine follow-ups send by rule, with checks on frequency, opt-outs, and approval logs.
- [Works Locally, Fails in Production? The Last Mile of Deployment](https://fxai.ai/en/notes/deploy-last-mile/): AI writes a site faster, but getting it live and stable still takes engineering discipline. A review of one real Cloudflare Pages launch and how we debugged it.
- [How to Deploy a Website to Cloudflare Pages for Free in Four Steps](https://fxai.ai/en/notes/deploy-website-to-cloudflare/): Connect a GitHub repo, set the build command, and add a custom domain to deploy a static site to Cloudflare Pages for free, with global CDN and free HTTPS.
- [How Should a Small Team Design Its First AI Coding Workflow?](https://fxai.ai/en/notes/design-first-ai-dev-workflow/): Move a small team from ad hoc AI use to a dev process where AI takes part in requirements, coding, testing, and review, with written rules everyone follows.
- [Engineering Guardrails for AI Coding: From Writing Code to Constraining It](https://fxai.ai/en/notes/engineering-guardrails-for-ai/): AI coding amplifies the engineering discipline you already have. Five guardrails keep output stable - behavior oracles, types, fast tests, traces, refactors.
- [Why Should an Enterprise AI Pilot Start With One Department?](https://fxai.ai/en/notes/enterprise-ai-pilot-vs-accounts-for-all/): Run an enterprise AI pilot in one department on one scored workflow with written go/stop criteria. Giving every employee an AI account first is not a pilot.
- [Enterprise Context Engineering: Making Company Knowledge Usable by AI Agents](https://fxai.ai/en/notes/enterprise-context-engineering/): AI agents stall in production when company knowledge conflicts or goes stale. Make context reliable, executable, and auditable, one workflow at a time.
- [GEO for B2B Export Websites: How to Attribute AI-Sourced Inquiries](https://fxai.ai/en/notes/export-site-geo-lead-attribution/): Track AI inquiries on a B2B export site in three places: a GA4 AI channel (the floor), hidden form fields, and self-reported source. Tag WhatsApp and email too.
- [From Coder to Agent Orchestrator: How AI Changes a Developer's Job](https://fxai.ai/en/notes/from-coder-to-orchestrator/): AI handles throughput and humans handle judgment. How developer work moves from writing code to system design, agent orchestration, guardrails, and review.
- [AI Agent Memory Architecture: Markdown Is the Source of Truth, the Database Is a Disposable Cache](https://fxai.ai/en/notes/gbrain-markdown-truth-source/): GBrain, Garry Tan's open-source agent memory system, keeps Markdown in Git as the only source of truth and treats its database as a rebuildable index.
- [Which GEO Metrics Should You Track?](https://fxai.ai/en/notes/geo-metrics-to-watch/): Track GEO in three layers (mention rate, citation rate, recommendation quality), check GA4 and Search Console for outcomes, and sample answers repeatedly.
- [Graph Engineering vs Loop Engineering: Graphs Sit on Top of Loops](https://fxai.ai/en/notes/graph-engineering-is-not-loop-engineering/): Graph Engineering sits one layer above Loop Engineering: a loop decides when one agent checks its work and stops; a graph decides which loop runs first.
- [Rolling Out Grok Bot at Work? Settle Access and Audit Settings First](https://fxai.ai/en/notes/grok-bot-enterprise-access-audit/): Before enabling Grok Bot company-wide, decide who gets access, how approvals are enforced, and what the Audit Log and Action Recording each actually capture.
- [Do Grok Bots Share Files? Isolating Bots on One Cloud Computer](https://fxai.ai/en/notes/grok-bot-shared-computer-isolation/): All Grok Bots on one Cursor account share one cloud computer, files and logins included. Each Bot's screen is a work surface, not a security boundary.
- [Grok Bot vs ChatGPT: A Cloud Teammate or Another Chat Window?](https://fxai.ai/en/notes/grok-bot-vs-chatgpt-cloud-teammate/): Since Aug 26, Grok Bot is in Cursor Pro, SuperGrok, and Cursor Teams. Unlike ChatGPT, it takes a task and keeps working on a cloud computer after you log off.
- [Brand Monitoring on X with Grok Bot: What the Daily Digest Should Say and When to Escalate](https://fxai.ai/en/notes/grok-bot-x-brand-monitoring-digest/): Hand X brand monitoring to a Bot. Each digest item gets a link, why it matters, and whether a person must decide. Write escalation rules before it starts.
- [From Idea to Product: How to Design an End-to-End AI Delivery Path](https://fxai.ai/en/notes/idea-to-product-delivery/): A delivery roadmap that takes an AI idea to a working product, with acceptance criteria at every step. Baize AI designs the path; your team builds the product.
- [AI Agents in IT Change Review: Which Steps Still Need a Human Sign-Off?](https://fxai.ai/en/notes/it-change-management-agent-signoff/): Agents can check IT change requests against configs, runbooks, and schedules. Release approval, change windows, and rollback owners still need human sign-off.
- [How to Write a Useful llms.txt File](https://fxai.ai/en/notes/llms-txt-how-to-write-useful/): llms.txt is a Markdown index at your site root (llmstxt.org v2, 2026). What to list, how to order links, how it differs from robots.txt, and how to test it.
- [MCP Security Risks: Which 4 Items Should You Review Before Connecting?](https://fxai.ai/en/notes/mcp-server-security-review/): Third-party MCP servers let agents call code on your network. Check the accounts the servers run under, input paths, authentication, version maintenance and credential rotation.
- [Robot Imitation Learning: Who Still Writes the Motion Script?](https://fxai.ai/en/notes/robot-video-imitation-vs-motion-script/): BeyondMimic tracks human motion on a real humanoid, from cartwheels to sprints. Waypoints, joystick control, and obstacle avoidance still need human-written costs.
- [Fine-Tuning vs. RAG: Connect the Enterprise Knowledge Base First](https://fxai.ai/en/notes/vertical-model-finetune-vs-enterprise-kb/): Fine-tuning a vertical model on weekly-changing facts bakes in stale data. Connect an enterprise knowledge base (RAG) first, then fine-tune for tone and format.
- [When Should a Voice Sales AI Agent Hand a Call to a Human?](https://fxai.ai/en/notes/voice-sales-agent-human-handoff/): A voice sales AI agent cannot take back what it says on a live call. Six handoff triggers, four facts to pass to the rep, and how to audit call recordings.
- [What Is AEO, and How Is It Different from Search Ranking?](https://fxai.ai/en/notes/what-is-aeo-vs-search-ranking/): A top search ranking does not mean AI Overviews, ChatGPT, or Perplexity will cite you. AEO tracks citations in AI answers. The missing step is the answer block.
- [What Is Loop Engineering? Make AI Check Its Work Before It Delivers](https://fxai.ai/en/notes/what-is-loop-engineering/): Loop Engineering makes AI check its work before it delivers: plan, execute, observe, correct, repeat. Covers the three loops and the templates they leave.
- [What Does a Secure LLM API Gateway Solve?](https://fxai.ai/en/notes/what-llm-gateway-solves/): An LLM API gateway gives all of a company's AI calls one entry point for authentication, routing, rate limiting, audit, model fallback, and cost logs.
- [Why Companies Shouldn't Just Buy Individual AI Accounts](https://fxai.ai/en/notes/why-not-just-buy-ai-accounts/): When every employee buys their own AI account, nobody can see who uses which model or what data leaves. Companies need one trusted relay or gateway instead.
- [Zero-Click Searches Are Rising. Do You Still Need SEO?](https://fxai.ai/en/notes/zero-click-search-still-need-seo/): About 68% of US Google searches ended without a click in early 2026. SEO still matters, but brand, local, and transactional KPIs should replace raw traffic.
- [Third-Party LLM Router Security: Why Cheap Proxies Put Agent Credentials at Risk](https://fxai.ai/en/notes/third-party-llm-router-security/): A practical guide to third-party LLM router risk, malicious model output, Agent Full Access, credential theft, sandboxing, and self-managed LLM gateways.
- [What Is Grok Bot? An AI Agent Guide to Persistent Teammates](https://fxai.ai/en/notes/grok-bot-persistent-ai-teammates/): What is Grok Bot? Learn how xAI's persistent AI agent uses a cloud computer, real Grok Bot use cases, Skills, Routines, Group Chat, and security approvals.
- [GLM 5.3 API for Codex, DeepSeek Harness, and Oh My Pi](https://fxai.ai/en/notes/baize-open-model-api/): GLM 5.3 is now available on Baize API. Configure the OpenAI-compatible API for Codex, DeepSeek Harness, and Oh My Pi. New accounts receive $10 in starter credit.
- [Run Qwen3.8-27B Locally on Mac with Unsloth](https://fxai.ai/en/notes/qwen38-mac-unsloth/): Pick an Unsloth Dynamic GGUF from Apple Silicon unified memory, then run Qwen3.8-27B with Desktop, Studio, or llama.cpp. On 32GB Macs, use UD-Q4_K_XL (~17.9GB).

## Optional

- [RSS feed](https://fxai.ai/en/rss.xml): Updates to the English notes
- [Sitemap](https://fxai.ai/sitemap-index.xml): Index of every crawlable page
- [Chinese site](https://fxai.ai/llms.txt): Chinese index, with every note in Chinese