Production-grade AI agents, with strong guardrails
I overhaul how your operations and support teams work with autonomous AI agents, governed by the same rules that keep 400+ critical SQL Server instances running: least privilege, blast-radius checks, change windows, human approval and rollback.
- ✓Reply within 24 h
- ✓Metadata-only access
- ✓Human in the loop
- ✓Access limited to metadata and metrics, no customer data
- ✓Guardrails and human approval on sensitive actions
- ✓Every action logged and reversible
Trusted by

01 · Security by design
Autonomous agents, never out of control
An AI agent is no more careful than the rules it is given. Every action passes the same seven gates a senior DBRE would apply, and autonomy is matched to risk: reads run freely, runbooks run pre-approved, production changes wait for a person.
- 01IntakeJira ticket triaged: P2, blocking over 90 s on the billing app.pass
- 02Diagnose (read-only)Waiting…pending
- 03Policy checkWaiting…pending
- 04Blast radiusWaiting…pending
- 05ApprovalWaiting…pending
- 06ExecuteWaiting…pending
- 07Verify & recordWaiting…pending
Read and diagnose: DMVs, wait stats, plans, msdb history, metrics.
Reversible actions from a tested runbook, on-call notified.
Production changes: the agent proposes, an authorised person decides.
Freezes, customer data, anything outside the written scope.
The production rules every agent inherits
- 01
Least privilege
Metadata and metrics only: DMVs, catalog views, msdb history. Your table contents and your customers' data are never read.
- 02
Written scope
Every agent has a declared role, an allow-list of tools and limits it cannot cross.
- 03
Blast radius first
Before acting, the agent sizes the impact: sessions, locks, replicas, rollback time.
- 04
Change windows and freezes
Changes follow your maintenance windows and freeze calendar. Outside them, the agent stages and waits.
- 05
Runbooks, dry run first
Actions go through tested runbooks that are safe to re-run, with a dry run before the real one.
- 06
Audit trail and rollback
Every action is timestamped, logged and delivered with the procedure to undo it.
- 07
SLO-aware stop
Agents watch your service levels. If an action degrades them, they stop, roll back and page a person.
- 08
Human in the loop
On sensitive actions, the agent proposes and an authorised person decides.
02 · Services
Agentic AI overhauls, with specialist consulting on call
The core engagement redesigns how your operations and support teams work, with AI agents doing the repeatable work under production guardrails. Data modernisation and SQL Server, my specialty, are available as consulting services on their own.
Rebuild your operations and support around AI agents, safely
From the first workshop to agents resolving real tickets in production: L1/L2 support, Jira triage, infrastructure and DBA diagnosis, inbound requests.
- Your processes reviewed and ranked by gain and by risk
- Agents in production, not mock-ups
- Security by design: metadata and metrics only, guardrails, human in the loop, rollback
- Documentation and handover to your team
- Start · 1 dayAI agents workshop€800 excl. VAT
- Build · fixed, 3 daysInbound requests agent€3,000 excl. VAT
- Extend · pack of 20 h (€2,000 excl. VAT)Hours bank€100 excl. VAT / h
Data modernisation and migration to Databricks
SQL Server estates to Azure, Kubernetes and Databricks: governance, cost control (FinOps, DBU) and pipelines ready for generative AI.
- SQL Server estates to Azure, Kubernetes and Databricks
- Data governance
- FinOps: DBU cost under control
- Pipelines ready for generative AI
Flash audit and SQL Server performance tuning
Wait stats, execution plans, indexing, locking: a prioritised diagnosis, fixes applied and skills transferred to your team.
- Wait stats, execution plans, indexing, locking
- Prioritised diagnosis
- Fixes applied
- Skills transferred to your team
03 · Case study
Samix HQ: a company run by AI agents
Samix Technology itself runs on a team of AI agents, Samix HQ. Each agent has a role (prospecting, bookkeeping, marketing, development, server operations), its own tools and its own limits.
They share a knowledge base, hand work to each other and ask for my approval before anything that commits the company: an email to a client, a purchase, a change in production. Every action is logged.
- PRProspectingAsks before any email to a clientapproval
- BKBookkeepingAsks before any purchaseapproval
- MKMarketingShares the knowledge base, hands work onscoped
- DVDevelopmentAsks before any change in productionapproval
- OPServer operationsAsks before any change in productionapproval

04 · Who does the work
A DBRE who builds agents, not the other way round
Evan Barke, 18 years of experience across databases, data platforms and operations. I have run the critical infrastructure of an estate of 400+ SQL Server instances (several petabytes) alongside Databricks, PostgreSQL and Elasticsearch, with an SRE approach and Infrastructure as Code. Today I build AI agents that work inside those environments, under control. Author of technical articles on SQLShack.
- 18 years
- of production experience
- 400+
- critical SQL Server instances
- 75%
- of tickets resolved by an AI agent
- 24 h
- to a first reply
CTOs and VPs of Engineering
You want AI agents in production without taking risks with security or with your data.
Data teams
You are moving a SQL Server platform to Azure and Databricks and want pipelines ready for AI.
Format
Engagements with scope and price agreed up front, remote or on site.
05 · References
What clients say
Software vendors, food groups, financial services: demanding environments, in France and abroad.
“Samix has been instrumental in diagnosing and improving performance across the park. The consultant has been reliable, knowledgeable and pragmatic in delivering the results we needed.”
“The real-time analytics platform Samix built on Azure, with streaming analytics from our greenhouses and data collection tablets feeding into a data lake, has transformed our BI capabilities and given us insights we never had before.”
“Samix created our master data universe, bringing together all our disparate data sources into one coherent system. It's been a game-changer for how we manage and leverage our data.”
06 · Resources
Latest articles
- Lifecycle and migration · 6 min
Windows Server 2016 end of support (January 12, 2027): what to do
On January 12, 2027, Windows Server 2016 leaves extended support. How to find the affected servers in 2 minutes, which option suits each one, and why to move SQL Server at the same time.
- Lifecycle and migration · 9 min
SQL Server 2016 end of support (July 14, 2026): what to do now
Since July 14, 2026, SQL Server 2016 no longer receives security fixes. What that changes in practice, what ESUs cost, and how to choose between SQL Server 2022, SQL Server 2025 and Azure.
- Performance · 10 min
Slow SQL Server: how to find the real cause before buying hardware
When an application slows down, the first reflex is often to add cores or memory. Here is the method we follow to find the real cause, with T-SQL queries you can run yourself.
Frequently asked questions
Which tasks can an AI agent take over?
L1/L2 support, Jira ticket triage, infrastructure and database incident diagnosis, and inbound requests (emails, quotes, invoices). We start from your queues and volumes to pick the tasks that pay back most.
How do you make an AI agent safe in production?
By design: the agent only reads metadata and metrics, never your customers' data. It has a written scope and limited tools, sensitive actions go through human approval, and every action is logged with its rollback procedure.
Which technologies do you build agents on?
Azure AI Foundry and Microsoft Agent Framework, and other models and frameworks to suit your environment. The choice depends on your cloud, your security rules and your existing tools (Jira, monitoring, databases).
How does an engagement work?
Each engagement has its scope and price agreed up front, remote or on site. Send your stack context and you get a reply within 24 hours.
Do you help with migrations to Databricks?
Yes: moving SQL Server estates to Azure, Kubernetes and Databricks, with data governance, cost control (FinOps, DBU consumption) and pipelines ready for generative AI.
Do you still offer SQL Server audits and tuning?
Yes, as specialist consulting: flash audit and high-volume performance tuning (wait stats, execution plans, indexing, locking), with a prioritised diagnosis, fixes applied and skills transferred. A free health score is a good place to start.
Do you work in French as well as English?
Yes, in both, with teams in France and across Europe.
Will you sign an NDA?
Yes, an NDA can be signed before any technical discussion.
Send your stack context
A few lines are enough: your tools, your volumes, what you want to automate or modernise. I reply within 24 hours.
Evan Barke, sole trader (EI), SIRET 914 474 358 00019