AI Weekly Digest: New Launches and the Hidden Threats of AI in Democracy
Week 41, 2025
This week’s stories show an industry continuing to jump into the mainstream. We’re reaching a critical point in time where we need to decide (and research) how AI will impact our political systems.
This week Nvidia improves reasoning models. OpenAI releases AgentKit, turning agent creation into a visual process. The Guardian raises alarms about AI’s long-term impact on democracy. And an article from Harvard reminds leaders that human judgment remains irreplaceable in the age of AI.
And we have a final shameless self-promotion easter egg🥚
Let’s start!
Nvidia Teaches LLMs to “Think” Before Speaking
Nvidia researchers introduced Reinforcement Learning Pre-Training (RLP), a method that rewards large models for generating internal reasoning chains before predicting text.
Stronger logic, fewer hallucinations, and better generalization without extra supervision is the result from their findings.
If proven scalable, RLP could mark a shift toward built-in reasoning during pre-training, not just during interaction with the users.
OpenAI’s AgentKit Turns Agents Into Workflows
OpenAI unveiled AgentKit, a visual development environment for AI agents.
Key parts:
- Agent Builder: drag-and-drop workflow canvas.
- Connector Registry: manage APIs and tools from one place.
- ChatKit: embed conversational agents in apps.
AgentKit signals OpenAI’s next phase: from API provider to AI operating system.
AI and Democracy: The Hidden Structural Threats
A Guardian op-ed argues that the danger of AI isn’t just election disinformation, but also institutional erosion.
The three slow-burn risks:
- AI replaces human communication channels.
- Control over algorithms consolidates political power.
- Automation reshapes journalism into a single narrative.
AI technology has the potential to reshape the foundations of democracy itself. We’re already seeing early signs, from the flood of AI-generated “slop” content influencing online discourse and muddying election narratives, but the deeper risks go far beyond misinformation.
Over time, AI could alter how citizens access information, how institutions make decisions, and who controls the flow of public dialogue. The real threat isn’t a single fake video or viral post; it’s the slow, structural shift of political power toward those who own and operate the algorithms that shape what society sees and believes.
HBR: What Great Leaders Need in the Age of AI
In their Harvard Business Review article, professors Herminia Ibarra and Michael G. Jacobides argue that thriving in the age of generative AI depends less on technology itself and more on leadership transformation.
They identify five key competencies that define effective leadership in this new era:
- Span organizational boundaries. AI fluency doesn’t come from reading reports; it comes from exposure. Leaders should build diverse networks that cross industries, regulators, and startups to understand how AI is being applied in practice.
- Redesign organizations. AI delivers returns only when companies rethink workflows and structures, not when they bolt automation onto legacy systems. True transformation means deciding where to automate, where to augment humans, and where human judgment must stay in control.
- Orchestrate team collaboration. The test isn’t just using AI: it’s integrating it into group decision-making. Leaders must choreograph how humans and algorithms think together, treating AI as an active teammate rather than a passive tool.
- Coach and develop talent. As AI changes the nature of work, employees need guidance to reskill and adapt. The best leaders replace inspection with coaching, creating a culture that rewards curiosity and experimentation over compliance.
- Lead by example. Adoption starts at the top. Leaders who use AI personally, testing tools, prompting models, exploring their limits, signal that experimentation is expected, not optional.
The authors close with a clear warning: AI won’t deliver value because firms spend on infrastructure, it will deliver value when leaders evolve.
Shameless Self-Promotion: Podcast of the Week
This week I joined Bitalk Podcast to talk a bit about AI! The podcast is in portuguese but you can access automatic english translated, if needed. Hope you like it!
Watch it here.
Final Thoughts
This has been another interesting week! Nvidia experiments with cognition. OpenAI lowers the barrier to creation. Policy thinkers warn about long-term governance. And leadership experts remind us that adaptation, not automation, defines success.
As always, stay curious,
– Ivo
