Sebastian Foo

Digital transformation consultant (World Bank, Deloitte, etc). Exploring how AI and other disruptive tech reshape governments, business and work.

[Sebastian Foo @SebastianFooPC]

10/10 PHP says critical security work rests on a small group of funded developers and volunteers. As AI expands discovery, resilience requires investment beyond detection: in maintainers, vendor patch obligations and tested recovery paths. Who is funding that capacity?

[Sebastian Foo @SebastianFooPC]

1/10 AI helped expose serious flaws. Patching them required complex engine changes and coordination with suppliers. Some services shut down as a precaution. Finding a bug and making systems safe are different tasks. 🧵

[Sebastian Foo @SebastianFooPC]

9/10 For government CIOs, ask: Which services rely on vulnerable PHP runtimes, including third-party products and managed hosting? Who owns each patch? Can it be verified? What happens if the supplier cannot deliver? “Our application is updated” is not enough.

[Sebastian Foo @SebastianFooPC]

5/5 How will the transition be managed? Potential long-term gains do not settle who bears adjustment costs or when. My takeaway: assess productivity, distribution, decision rights and transition capacity together.

Research agenda: https://t.co/Tz84f5VMfl

[Sebastian Foo @SebastianFooPC]

1/5 AI's economic impact cannot be assessed through productivity alone. Who receives the gains, what constrains deployment, and who controls decisions also matter. Four questions prompted by Brynjolfsson, Korinek and Agrawal's research agenda.

[Sebastian Foo @SebastianFooPC]

4/5 Who retains authority? Delegating a task to AI raises questions about who sets objectives, authorises actions and can challenge outcomes. My implication for public services: assess the allocation of authority alongside efficiency.

[Sebastian Foo @SebastianFooPC]

3/3 Yes, manual planning is time-consuming and imperfect. Yes, AI might reduce omissions and free up staff time. But it must also depend on consultation, decision quality, and actual service delivery, not on the number of plans generated or how quickly one is created.

[Sebastian Foo @SebastianFooPC]

1/3 AI may help draft an education plan faster. But the purpose of the plan isn't paperwork. It's to secure appropriate support for schools, teachers, and students. If schools automate this work, the real test is what happens after they approve the draft.🧵

[Sebastian Foo @SebastianFooPC]

2/3 Three tests matter: Is the plan accurate and specific to the school, to the teacher, and to the child? Did the plan involve parents to make evidence-informed decisions? Were the agreed services delivered? Measure all three before claiming the technology improves support.

[Sebastian Foo @SebastianFooPC]

A UK trial found an AI lesson-planning assistant saved primary teachers about 49 minutes a week. Yet teachers used it for only 7% of lessons on average, and usage declined over 10 weeks. The question: Will the savings hold up in typical classroom workflows?