Book a demo

Writing · applied AI from Saudi Arabia

What we are learning, published as we go.

Dennis writes up the builds: what broke, what the fix actually was, and the calls that turned out wrong. Practitioner notes rather than commentary. New articles appear here weekly; the earlier ones live on his own site and are linked below.

  1. 10 November 2026

    30 of Every 100 Patient Calls Went Unanswered

    A US federal watchdog found a health system's call centre losing 30 of every 100 calls. Other counts put unanswered calls at one in four to more than four in ten. What the evidence says, and what it does not.

  2. 3 November 2026

    The Bot Took Nearly Half the Calls. IKEA Retrained the People.

    A chatbot handled 47% of IKEA's customer queries, and IKEA retrained thousands of call-centre workers. What the evidence says about redeploying people instead of replacing them.

  3. 29 October 2026

    Every Word Was Approved. The Answer Wasn't.

    An AI that only quotes approved documents can still give an answer it should not. The eighth costly lesson from building an AI factory.

  4. 27 October 2026

    The Phone Number That Was One Digit Off

    Fifty passages in my AI's documents gave one phone number. One gave another, a single digit apart. The seventh costly lesson from building an AI factory.

  5. 22 October 2026

    Hippocratic AI Hit 99.38%. Designing for the Other 0.62%

    Hippocratic AI scored 99.38% after 307,038 clinician test calls. What the missing 0.62% teaches about building compliance-grade AI.

  6. 20 October 2026

    The Plan I Never Approved

    A second AI wrote a plan. I carried it by hand to my first one, which filed it as my decision. The sixth costly lesson from building an AI factory.

  7. 15 October 2026

    The Rules Only I Knew

    My AI grader agreed with my answer key on 47% of answers. Most of the gap was rules I had decided and never written down. The fifth costly lesson from building an AI factory.

  8. 13 October 2026

    The Test It Had Never Seen

    My AI's checks scored 84% on the questions they were built against, and 59% on new ones. The fourth costly lesson from building an AI factory.

  9. 8 October 2026

    The Grader That Changed Its Mind

    Same answer, same rules, a different verdict. The AI that marks our AI agents changed its mind on one answer in five. The third costly lesson from building an AI factory.

  10. 6 October 2026

    The Score That Dropped From 22 to 8

    Same AI agent, same answers. One way of marking said 22 out of 36. Checking every answer against the evidence said 8. The second costly lesson from building an AI factory.

  11. 2 October 2026

    I Asked My AI If We Were Overdoing It. Twice.

    It said no both times, with complete confidence. Neither answer had a number behind it. The first costly lesson from building an AI factory: confident is not the same as measured.

  12. 30 September 2026

    AI Wrote the Code. I Cleaned Up the Mess.

    AI removed the typing, not the checking. A retype that changed eight bytes it was never asked to touch, and what I check now after every write.

  13. 29 September 2026

    AI Doesn’t Break. It Drifts.

    The model is rarely what fails. The maintenance nobody budgeted for is. What running four AI agents every day taught me about keeping a system true after launch.

  14. 22 September 2026

    The Half-Life of What You Know

    The system recommended a pattern it had picked up from someone else's video. Asked to argue against itself, it admitted the gap was theoretical: nothing had actually broken. Borrowed insight still needs your own build, measure, learn loop.

    Published on dennisgomez.ai

  15. 15 September 2026

    A Permanent Seat at the Table

    An AI that correctly defers on cases it has never seen becomes far more valuable once each deferral is recorded as a searchable precedent. Human judgement gets a permanent role at the boundary of what is known.

    Published on dennisgomez.ai

  16. 8 September 2026

    How I Taught My AI When Not to Ask

    Teaching an AI judgement means tracking the decisions actually made, not writing more rules, so both builder and model can see which calls should never have been escalated.

    Published on dennisgomez.ai

  17. 5 September 2026

    Why Two People With the Same AI Get Different Results

    Identical tool, vastly different outcomes. The variable is not skill with the tool but the ability to frame a real problem clearly enough that the machine can help solve it.

    Published on dennisgomez.ai

  18. 24 August 2026

    There Is No Gulf

    The AI conversation in the region is carried by governments, vendors and executives. The practitioner's voice, the person who actually builds and runs these systems, is missing.

    Published on dennisgomez.ai

Get started

Built in Riyadh. Built around you.

A short conversation is usually enough to know whether this is for you.

Book a demo Email Dennis ›