The Future We Live In, Amongst Agentic AI

Using AI's impact on the Communications field as a Case Study

Hi All,

I recently joined a panel to discuss the intersection of AI and communications at CityU University in Hong Kong, alongside industry experts with diverse fields of expertise—a software developer, a lawyer, an educator, a policy analyst, and others.

One of the key points I brought up was the framework I walked everyone through, which outlined how to think about AI and its impact on the future of the professional field of communications. The focus is not on how AI is currently affecting our way of communicating and working, but rather on how agentic AI will disrupt our thinking about knowledge work as a whole.

AI x Communications

Let me break down my framework.

Phase 1: Distribution as a Bottleneck

Twenty to thirty years ago, the bottleneck for companies to convey their message to the public or any stakeholder was primarily in distribution. If we use Hong Kong as an example, and I wanted to obtain information about my company’s performance for Wall Street, aside from direct channels of communication or through bankers. I would likely need to know the reporters at The Wall Street Journal or The New York Times who cover Asia.

Now, newspaper real estate was truly scarce as well, so as the company, you’d have to fight to sell the most interesting story, or what the reporter then deemed as the most important story for people sitting on the other side of the world to know. The distribution was the true bottleneck for information dissemination.

Phase 2: Rise of Digital Media

Then, let’s say about 10-15 years ago, we saw the peak of digital media platforms like HuffPost, BuzzFeed, Vox, Vice, and so on. They were able to cater to a broader audience, often younger than the traditional outlets. They carved out a niche for each outlet, and in some ways, you can say “lowered the standard” for stories to make it in a publication. Not to say these stories were less worth knowing it’s just that there was a lot more real estate for more stories to be told. However, what it led to was an overabundance of information and content, and the creation of what we now know as echo chambers of worldviews. Their business models couldn't be sustained, and now most of these platforms have shut down.

Phase 3: Rise of False Information and Need for Governance

Then we saw the rise of social media and the internet platforms themselves becoming news sources for the majority. Misinformation and disinformation became rampant, but on the other hand, there was a rise in citizen journalists. Information access was no longer an issue, and distribution no longer had a high barrier to entry. Expertise and the need to dissect information became scarce. And throughout these times, PR agencies revamped themselves to follow the trend, from helping companies access journalists and vice versa, connecting reporters to potential story leads, to then focusing on building out relationships and expertise with “trade media” or “regional media,” to then pivoting from pure corporate messaging to offering “digital solutions.”

The other big issue that arose from the democratization of information dissemination is the increasing number of false information and misinformation, and a recent study by the Reuters Institute at Oxford University showed that there was an enormous discontent from news consumers regarding the rampant spread of misinformation; however, most believe it is the responsibility of platforms to monitor, instead of the government’s. However, each company is led by a person and/or sponsored by a pool of capital backed by a party, which has led to the extreme political polarization of media and tech platforms.

Phase 4: AIGC Floods the Internet

This leads us to the current phase, where we’ve seen major tech companies implement content moderation processes, including both human content moderators and AI-powered moderation tools. However, on the other hand, content creators, whether professional or casual users, have also adopted AI for content creation. While there is no definitive percentage of the amount of content online that is generated by AI, some studies suggest that nearly 57% of all online text content is AI-generated or translated. An Ahrefs study found that as of early 2025, ~74% of new webpages included some AI content. And it’s not just the normie consumers, even the media experts sometimes cannot tell the difference between AIGC and real photography/ video/ content. (Much, much different from when I first covered deepfakes in 2019 - oh god, I am so embarrassed to link this, but here is a video of baby reporter me)

And in my own estimate, I think we can easily see AI automating at least 30% of the work that writers, researchers, and PR professionals took hours to do within minutes. I’m not saying they are at the top-notchh quality, but they can do the work. So, suddenly, media monitoring, interview briefings, research packages, press releases, callnotes and emails can be completed within minutes. We’re seeing that the ability towrite is no longer as vital, but to write well and write to influence is more critical than ever. However, to achieve that and avoid appearing like generic AI writing, industry expertise is necessary. We need more professionals who previously relied heavily on their writing capabilities to be now able to carve out their expertise. If you work with tech companies, you should have a basic understanding of technology. If you work with consumers, you need to know consumer companies inside and out. The execution bit is less difficult to accomplish, the strategic and critical thinking bit is more important, and that leaves many junior staff in an awkward position.

Phase 5: Coworking with Agentic AI

However, we then look forward to an Agentic AI phase, where, in 2-3 years or even less time, we may have agents embedded in customer service (if not already) and sophisticated work processes. Then, what will be vital is the ability to prompt and manage agents to complete tasks correctly. Additionally, it will be essential to train the agents to respond to different stakeholders and each individual interacting with them in a customized and tailored fashion. And more importantly, people who have industry expertise and can command AI tools.

In essence, we’ll need more “operators” or “project managers” and even “insight leaders” but fewer “executioners.” What we’ll likely see in this industry is the use of agentic AI to eliminate toil and keep humans in control for narrative, risk, and trust.

---

The Future We Live In, Amongst Agentic AI

So that leads me to the bigger topic of today, the future of coworking with Agentic AI.

To start, this is how the smart smart researchers at Bernstein defined Agentic AI:

Agentic AI is the space that translates the LLM’s immense potential into autonomous action. The expectations from AI have already shifted to be outcome-driven - performing multistep, complex tasks like scheduling/prioritizing patients basis urgency after a conversation, booking tickets or assisting radiologists with diagnosis.

This ability to act, learn and adapt going beyond an LLM response is what makes Agentic AI the true frontier to watch out for, if one were to assess disruption due to AI.

Awaiting the Big Breakthrough

Agentic AI has been the talk of the town for the last 2 months. Every company is saying, “We have agentic AI". However, most layman consumers have not been “wowed” the same way when even the first ChatGPT consumer-facing app came out in late 2022.

The reason is well, for one, we humans have high expecations and become harder and harder to satisfy when fed with beter things - think about your taste buds, I took my husband back to my favorite thai food hole in the wall from my uni days and honestly it tasted so bad I couldnt even finish my meal, because since then I’ve been fed better food and my taste buds have become more and more snobby.

And two, we haven’t really figured out how to use agentic AI to really co-work/ collaborate/ unleash its potential yet.

How to understand the relationship between agentic AI and generative AI? Think of generative AI as the “brain” and the agentic AI as the “execution.”

An example, whether it’s your ChatGPT or your DeepSeek, it was able to plan you a perfect 5-day itinerary for Osaka that is toddler-friendly and combines scenery, history, and modern city culture. That is what the Generative AI could do for you. What the Agentic AI can now do is actually help you book the car, book the restaurant, and the tickets for the historical sites. (like a personal assistant)

For most, the agentic AI sphere is currently restricted to the human assistant stage in its evolution, according to Bernstein: “it automates most of the mundane back-end tasks, handles low-stakes customer conversations, and aids an expert for high-value tasks with human touchpoints. It is not replacing humans across the value chain, but rather unevenly - highest impact is in jobs like data entry, HR/IT support, administrative staff, junior sales representatives/software engineers, and customer support.”

That said, while it has not become fully autonomous, it does threaten a major disruption in certain IT, Customer Service, and Sales roles.

In Bernstein’s recent report, AI vs. Human: Agentic AI: What will a Billion Agents change? 100 Startups show the way, the researchers tried out 100 Agentic AI apps, and they summarized that agentic AI’s current evolution is best understood as three practical “leaps” that map neatly onto knowledge work:

1. Truncating the Toil – automating routine, high‑volume tasks (data entry, simple inquiries).

2. Structuring the Unstructured – ingesting long documents/web pages/calls/news releases and turning them into queryable, structured data.

3. Assisted expertise (human‑in‑the‑loop) – tackling more complex tasks (drafting, diagnosis, outreach) with explicit human checkpoints.

The conclusion?

The review of ~100 agent startups shows most systems are still in “human assistant” mode, automating back‑office toil and aiding experts, rather than replacing entire workflows end‑to‑end.

And earlier, I pulled a number out of a hat, saying that AI could replace ~30% of knowledge workers' tasks right now, well, McKinsey (probably more scientific in their methods than mine) estimated that to be much higher. In fact, McKinsey estimates current gen‑AI can technically automate activities that absorb 60–70% of employees’ time, so agentic AI can be assumed to actually complete even more tasks and preserve employees’ time.

And plugging in this interview with Diana Wu David, who is the Director of Futures at Service Now, where we talk about the future of work. I think it’s important to remind us all that, instead of thinking about job replacement, let’s think about how to manage and control AI better.

---

So... what I’m trying to say is, (and responding to some people upset about how companies will always be future proof and we humans get laid off), we should not be limited to thinking the only way to be “future proof” is to stay at an organization or a role forever. Instead of staying future-proof, being adaptable to the future workforce. Organizations take months or years to implement new processes and strategies. But for each individual, it’s a personal choice and one that can be made much more swiftly.

If my father’s generation hadn’t learned to use Microsoft Office or how to better leverage the internet to communicate more efficiently, he wouldn’t have remained relevant in the workforce. Environments evolve, societies change, and the companies that stay relevant also change. It takes one’s effort to change with the times, too. In the mid-term at least, it’s not “AI replacing us,” it’s people who can leverage AI to better optimize their existing workflow/ demand that will “replace us” (of those who are complacent to learn and adapt).

​​The Agentic breakthrough has not come yet, but it is coming.

---

I know there has been a TON of AI (LLM) news from DeepSeek, Alibaba, and Tencent. I will provide a deeper analysis later in the month once the dust has settled. And coming to you on Thursday is an episode of Differentiated Understanding with an AI investor on Neoclouds, the VC bubble, and the capital race in AI. Stay tuned.

---