From the blog
Notes from our engineers and consultants on AI, software and digital transformation in the Kingdom.
AI Voice Agents in Saudi Arabia: Nine Tests Before You Answer Calls
Before an AI voice agent answers real customers, test Arabic dialects, interruptions, booking accuracy, safe tool access, human handoff, privacy, outages and operating costs.
Passkeys or OTP? A Practical Login Upgrade Plan for Saudi Businesses
Passkeys can reduce phishing risk and simplify sign-in, but a safe rollout needs staged enrollment, strong recovery, step-up controls and clear policies for customer and privileged accounts.
Webhook, Queue or Polling? How to Build Reliable CRM–ERP Integrations
For business-critical integrations, receive changes with webhooks when available, acknowledge quickly, queue work asynchronously, make side effects idempotent, retry transient failures with backoff and jitter, and isolate persistent failures for investigation.
MCP or Direct API? How Saudi Businesses Should Connect AI Agents to Internal Tools
MCP can standardize how AI agents reach business tools, but it does not replace access control. Choose MCP or direct APIs by integration scope, expose only necessary tools, keep approvals on sensitive writes, and map data flows before production.
Should AI Coding Agents Be Allowed to Merge Code? A Control Model for Saudi Teams
AI coding agents can handle real engineering work, but production authority should expand gradually. Use isolated branches, deterministic tests and security scans, human review for material changes, and existing CI/CD gates before production deployment.
What Should an AI Agent Audit Trail Record? A Practical Guide for Saudi Businesses
An AI agent audit trail should record who initiated the run, what the agent was asked to do, which tools and data it accessed, the actions it attempted, approvals, results, errors and final outcome—without automatically storing every prompt or personal-data field.
Can You Trust Your AI Assistant’s Sources? A Bilingual RAG Test
A practical guide to testing Arabic and English AI assistants: check retrieval, source citations, outdated documents and when the system should say it cannot answer.
Where Arabic LLMs are ready for production — and where they are not
A field guide to what actually works today in Arabic-language AI: retrieval, drafting, summarisation — and the places we still keep a human in the loop.
A practical data checklist before your first AI project
Five questions we ask every client before writing a line of AI code — and what the answers tell us about project risk.
Automation that respects the way your teams already work
The best automation disappears into existing habits. The worst demands new ones. How we decide what to automate — and what to leave alone.