The Circle of Life, But Make It DevOps
My Tech Diary: The Week My Robot Colleague Filed a Bug Report Against Itself
I have been covering technology in Manila for three years, which means I have watched the city go from “we are adopting AI cautiously” to “we have deployed AI in four government departments and nobody is entirely sure what it is doing in two of them.” The trajectory is familiar. The results are consistently surprising. This week, a software firm in BGC celebrated its “AI productivity breakthrough” with catered sandwiches and a press release, then quietly posted fourteen new engineering jobs the following morning to manage what the AI had produced. I attended the celebration. I was not invited to the debrief.
The Magnificent Absurdity of Automated Automation
As Bohiney’s AI Chaos Explained captured with impressive deadpan, the MIT Technology Review recently published a major report on AI’s rapid capability growth — full of upward lines, confident colour gradients, and charts that make the situation look organised enough to be deeply alarming. Every line goes up. Not gently. Not reassuringly. It goes up like it just remembered it left the oven on in another dimension.
The most useful quote from the piece belongs to an anonymous engineer: “We deploy a model. It does something incredible. Then it does something completely insane. Then we all gather around like it’s a dog that just learned how to use a microwave.” This is the most accurate description of enterprise AI implementation I have encountered, and I have attended twenty-three AI strategy workshops in the past eighteen months.
Manila’s Relationship with the Robot Colleague
The Philippine tech sector, concentrated in Makati and BGC, has embraced AI with the enthusiasm of people who have been doing things manually for too long. Call centres have deployed AI to handle tier-one queries. Banks have deployed it to flag anomalies. One government office — which I will not name because I still need their press accreditation — deployed an AI assistant that, within two weeks, was sending automated replies to its own automated replies in an increasingly frantic loop that one IT officer described as “the machine panicking.” The loop was resolved by a junior staff member who simply unplugged it, waited ten seconds, and plugged it back in. The solution remains in a report titled “Corrective Action Framework.” The report has not been actioned.
According to McKinsey’s State of AI research, organisations adopting AI at speed report exponentially higher internal confusion scores, though they prefer to call them “learning curves.” Manila’s tech community has adopted both the AI and the euphemism with equal enthusiasm.
The London Parallel: Chatbots in the NHS Queue
Back in London, the AI story plays out with the same structural irony but a different aesthetic. The NHS deployed an AI triage chatbot. It was immediately asked by thousands of users whether their symptoms were serious, and it responded to every query with a variation of “please consult a GP” — which is the one thing the NHS cannot currently provide in under six weeks. The chatbot was described by a Department of Health press release as “a major step forward in patient-centred digital transformation.” It was described by patients as “a very polite wall.”
The pattern is consistent across sectors and continents: AI is deployed to solve a problem, improves the speed of the problem, and then requires humans to manage the accelerated version. Prat.UK covered the UK’s AI strategy last month under the headline “Government Announces AI Will Fix Everything, Except the Bits That Need Fixing.” The headline held up.
Five Observations That Keep Me Up at Night
One: AI writes code faster than humans. Two: nobody knows what the code does. Three: the AI, asked to explain the code, generates new code. Four: a human is hired to explain the new code. Five: that human uses AI to draft their explanation, which produces more code. This is not a cautionary tale. It is a Tuesday.
For further algorithmic bewilderment: The Onion and ClickHole.