Microsoft's AI Chief Says Your Job Ends in 18 Months. SQL Server 2025 Disagrees

This week, Microsoft AI CEO Mustafa Suleyman told the Financial Times that most tasks that involve “sitting down at a computer” will be fully automated by AI within the next year or 18 months.   As “compute” advances, he said, models will be able to code better than most human coders.

This is the same Microsoft that, three months ago, shipped SQL Server 2025 — the most feature-dense release in a decade. A release that added an entirely new category of things that DBAs, developers, and data engineers need to learn, manage, secure, and troubleshoot.

Same company. I wonder if the left hand knows what the right hand is doing.

Let's Take Him at His Word

Suleyman says AI will automate the work of anyone sitting at a computer. SQL Server 2025 just shipped the following — every one of which requires a human being sitting at a computer:

Native vector data type and DiskANN indexing. Your developers are going to start storing embeddings in SQL Server. Someone needs to size the indexes, figure out why VECTOR_SEARCH is slow, and explain to management why the new 'AI column' doubled the storage footprint. That someone is you.

CREATE EXTERNAL MODEL. You can now define and call AI models from inside the engine using T-SQL. Which means someone has to govern which models are callable, manage the creds, handle endpoint failures, and explain to the security team why SQL Server is making outbound HTTPS calls to Azure OpenAI. That someone is also you.

AI_GENERATE_EMBEDDINGS and AI_GENERATE_CHUNKS. Building RAG patterns inside the database engine. This is compute-intensive work running on your SQL Server — the same SQL Server hosting your OLTP workloads. Who is configuring Resource Governor to keep it from stomping production?  Still you.

Change Event Streaming. Real-time data streaming from SQL Server to Azure Event Hubs, built on the transaction log. When the log blows up because nobody tuned the throughput settings, the 3 AM page goes to a human.

sp_invoke_external_rest_endpoint. The database engine can now call any HTTPS endpoint. Directly. From T-SQL. This one deserves a closer look.

Your SQL Server Is Calling the Internet Now

This is the single most significant new attack surface, integration point, and troubleshooting headache that shipped in SQL Server 2025. It is also, genuinely, the most powerful.

The feature is disabled by default. Here is how to turn it on:

EXEC sp_configure 'external rest endpoint enabled', 1;
RECONFIGURE WITH OVERRIDE;

And you will need to grant the permission explicitly:

GRANT EXECUTE ANY EXTERNAL ENDPOINT TO [YourAppLogin];

Here is what a basic call looks like. This hits a public API and returns the result as JSON:

DECLARE @ret INT;
DECLARE @response NVARCHAR(MAX);

EXEC @ret = sp_invoke_external_rest_endpoint
    @url     = N'https://httpbin.org/get',
    @method  = 'GET',
    @response = @response OUTPUT;

SELECT 
    @ret AS return_code,
    JSON_VALUE(@response, '$.result.origin') AS caller_ip;

That works. It is clean. It is powerful. And it just made your SQL Server a REST client.

Now here is what happens when the endpoint is slow:

-- Timeout defaults to 30 seconds.
-- Your transaction is waiting on an external service.
-- Your locks are held.
-- Your users are waiting.
-- Nobody told the DBA this was in the stored procedure.

EXEC @ret = sp_invoke_external_rest_endpoint
    @url     = N'https://some-slow-vendor-api.com/v1/data',
    @method  = 'POST',
    @payload = @requestBody,
    @timeout = 120,  -- two minutes, holding locks
    @response = @response OUTPUT;

When that vendor API goes down at 2 PM on a Tuesday and every transaction in the application starts stacking up behind it — that is not an AI problem. That is a DBA problem. You will be the one tracing it in sys.dm_exec_requests, finding the blocking chain, and explaining to the application team why calling a third-party API inside a transaction may not be the best idea.

This feature also requires HTTPS with a certificate trusted by the host OS, uses DATABASE SCOPED CREDENTIAL for authentication, and needs its own security review for every endpoint your developers want to call. That is governance work. Human governance work.

The Math

Suleyman says 18 months until the work is gone. His own product team just shipped five major feature categories — each one creating new DBA work that didn't exist six months ago. New data types to understand. New indexes to tune. New external dependencies to monitor. New security surface area to lock down. New resource consumption to govern.

The robots are not taking the jobs. They are becoming the job.

More to Read:

Fortune: Microsoft AI chief gives it 18 months — for all white-collar work to be automated by AI
Microsoft Learn: sp_invoke_external_rest_endpoint (Transact-SQL)
Microsoft Learn: What's New in SQL Server 2025
Microsoft Learn: Vector Search and Vector Index in SQL Server
Microsoft: SQL Server 2025 is Now Generally Available