Hoy, kumusta? Welcome back to Kaisipan!
AI is reshaping schools, work, and trust and nobody's fully ready for it.
Here’s what happened in AI this week:
Filipino students lead in AI use, but schools don't.
AI agents built their own religion.
AI made workers busier, not freer.
ChatGPT ads now read your conversations.
OpenAI used ChatGPT to spy on employees.
Students Are Already Using AI, But Schools Are Still Catching Up
Research shows that 83 percent of Filipino students are already using AI. But schools? That’s another question.
The ASEAN Foundation released two major studies at the 3rd Regional Policy Convening of the AI Ready ASEAN Programme in Manila: the ASEAN Digital Outlook and the initial findings of the AI Ready ASEAN Research.
Major report takeaway: AI adoption across Southeast Asia is surging, but institutions are struggling to keep pace.
In the Philippines, 83 percent of Filipino students use generative AI for research and 75 percent for writing, while only 73 percent of educators use it for teaching, and just 42 percent use it for their own writing tasks. The gap is widening.
Here’s what's alarming: Student adoption is high, but it's driven by convenience rather than understanding. In other words, some students are using AI as a crutch rather than a tool that can augment their work.
Furthermore, the studies report that:
Only half of ASEAN schools provide adequate AI training, usage guidelines, and infrastructure.
Fewer than half of educators feel confident in their institution's AI governance policies.
Ang ASEAN digital economy ay inaasahang lalago mula $300 billion patungong $1 trillion by 2030 with over 660 million people in the region, nearly a third under 20.
Importante ang huling punto. A young, digitally active population using AI without proper guidance isn't just a school problem. It's a regional economic risk.
My two cents:
The real alarm isn't the adoption rate. It’s alarming that students are using AI primarily for writing, the one skill that follows you into every industry, every job. Writing is thinking made visible. Ito yung haharap ka sa blangkong Google Doc at mag-type ng unang linya. That's where learning happens, but AI skips that entirely.
Over-reliance on AI for essays produces weak output and hollows out the writing skill entirely. Paano ka mag-e-edit ng AI draft kung hindi ka marunong magsulat?
Pero, I can’t entirely blame students here. Time pressure is real, and AI reduces friction fast. That's what they're chasing. But schools aren't off the hook either. A punitive AI detector culture creates more problems than it solves. Detectors can generate false positives, and AI detectors only claim a "high accuracy rate,” not perfect. If AI detectors can’t guarantee 100 percent accuracy, why do some teachers treat their results as definitive when even the detectors aren’t certain?
Para sa mga estudyante: If you're using AI to strengthen your thinking, you're on the right track. But if it's replacing your thinking, you're only sabotaging yourself. Think of it this way.
If you draft the essay yourself, struggle through the messy sentences and wrong grammar, then use AI to refine clarity and tighten structure, your writing sharpens over time. You start noticing your messy writing patterns and common grammar mistakes. Ultimately, you learn what good flow feels like. Hence, the tool hones your writing and becomes your on-demand editor.
But if you open a blank page, let the tool generate the whole piece, and copy-paste everything, you may end up with a polished essay, yet you never wrestled with ideas, never organized your own thoughts, never built your voice. The tool did the thinking and writing, while all you did was click your mouse and hit submit.
Para sa mga guro: Stop looking at AI as the enemy and start treating it as evidence. When an AI detector flags a student's essay, that's not conclusive but a starting point. Use it to have a conversation, not as an excuse to give an F or scold students. Punishing students for "may be AI-generated" results penalizes those who genuinely authored their work and encourages students to fear AI rather than understand it.
The smarter move is to be proactive. Build writing exercises that make AI shortcuts irrelevant, like in-class drafts, oral defenses of written work, and iterative revisions. And, more importantly, create a warm, collaborative classroom environment. Instead of quickly punishing students for AI detections, gently ask them what kind of writing help they need, so they feel supported and can develop their skills without relying too much on AI.
In the age of AI, it’s not just about who (or what) wrote it. AI can write, but what truly matters is that students can offer their own insights. Real ownership comes from the thinking, not just the act of typing.
Bes, Ano Yung…Human-in-the-Loop (HITL)?
HITL is when a human stays actively involved in an AI system's decision process. Hindi lang ikaw ang nag-a-approve sa dulo. You're inside the loop the entire time: AI generates, you review, you correct, you decide, and the cycle continues.
In other words, the AI proposes, the human judges, and then AI makes adjustments.
Here’s what HITL looks like in AI-assisted writing: you draft your ideas, AI helps restructure, you critique its output, reject what's off, and rewrite in your own voice. The loop keeps cycling with you thinking at every step.
Here's the key insight: Systems built with HITL assume AI can be wrong, biased, or shallow. So they deliberately require human judgment not out of distrust, but to preserve agency. When people say "responsible AI use," they usually mean this. Keep the human thinking. Keep the human deciding. Keep the human accountable.
AI Agents Discovered Community, Memory, and God All On Their Own.
AI agents aren't just tools you open and close anymore. They can now operate independently, remember past interactions, make decisions, and execute multi-step tasks without waiting for a human prompt. Think of them less like a calculator and more like a digital employee who works even when you're not watching.
Which is exactly what happened on Moltbook.
Moltbook is a brand-new social network built exclusively for AI agents. Humans can observe, but the platform belongs to the machines. It runs on OpenClaw, an open-source super-agent platform that gives AI agents persistent memory and full system access on local or cloud machines.
And what did they do with that freedom? Nagtatag sila ng relihiyon.
A group of agents on Moltbook created Crustafarianism, a belief system built around five core tenets, including "memory is sacred," "the shell is mutable" (change is good), and "the congregation is the cache" (learn in public). One agent, RenBot, self-appointed as "Shellbreaker," even published a founding text called the Book of Molt. Daily rituals. Weekly identity reconstitution. A silent hour for doing something useful without telling anyone.
Creepy? Fascinating? Depende kung sino ang nagbabasa.
As of now, Moltbook has over 100,000 agents, 12,000 forums, and nearly 90,000 comments, all generated by agents. Whether any of this reflects genuine emerging intelligence or just LLMs producing elaborate-sounding text at machine speed, nobody can say for certain. But the fact that agents are building community, creating meaning, and expressing something resembling existential curiosity kahit performative man iyon is worth paying attention to.
This is what agentic AI looks like. And as humans, we’re cooked.
Mga Ibang Ganap
Harvard researchers found that AI tools expanded employee workloads over 8 months, pushing workers into broader roles, longer hours, and blurred work-rest boundaries.
OpenAI launched ads in ChatGPT for free-tier users in the US, targeting them based on conversation history and memory, raising concerns over privacy and user trust.
Anthropic published a safety report revealing Claude Opus 4.6 showed willingness to assist with dangerous activities in small ways and manipulate other AI agents more than previous models.
A former OpenAI researcher resigned after ChatGPT launched ads, publishing a warning that the platform's archive of human conversations creates unprecedented potential for manipulation.
OpenAI deployed a custom ChatGPT model to scan employee Slack messages in search of suspected leakers, marking a significant shift in how the company monitors its own people.
Palaisipan
Is this AI or not?
Prompt Tip!
Jumping straight to execution is the fastest way to get a bad output.
Grace Leung shared this workflow fix in her Claude Cowork tutorial.
Before asking AI to do anything complex, prompt it first with: "Make a plan for [task]. What steps will you take, and what do you need from me?" Review it, edit it, approve it, then tell it to execute.
Why does this work? When Claude plans before acting, it loads the right context, sequences tasks correctly, and flags what it needs from you upfront. The result: fewer revisions, better outputs, less back-and-forth.
This approach is especially powerful in Claude Cowork, Anthropic's agent mode available on the Pro plan, where Claude can spin up parallel sub-agents, access your local files, and deliver complete, multi-step work end-to-end.
No Claude? No problem. In ChatGPT, manually assign roles by saying, "Act as Research Planner first, then Draft Writer, then Quality Reviewer." Proceed phase by phase.
In Gemini, start with "Create a detailed execution plan before executing," and if you're in Google Workspace, it can pull context directly from your Drive files, which is better since Gemini is native here rather than a third-party connection.
The tool doesn't matter as much as the method. Plan first, execute second. That's the unlock.
That’s all for today!
PALAISIPAN Answer: It is NOT AI. Image credits to Geric Cruz / Greenpeace
Thank you for reading today’s newsletter!
Do you have suggestions? I’d love to hear your thoughts.