When Your CTO Joins the Competition: Mesolitica's "Future is Solo" Pivot and the AI Market's Brutal Reality

I spent 1 hour 45 minutes in November with Khalil Nooh from Mesolitica (or MaLLaM, depending on which part of their open-source journey you caught them at), and the conversation felt like watching a startup's entire lifecycle compressed into a single coffee chat. The bottom line: Mesolitica's story isn't just about a Malaysian AI startup navigating technical challenges—it's a live case study of what happens when AI hype collides with market reality, and how the smartest founders pivot from "we'll build the future" to "let's teach everyone to build their own."

The conversation started with Husein's departure. Not the usual "co-founder moves on" story, but something more tangled: he took a corporate role at Scicom MSC Berhad as Senior Manager of AI Research & Engineering (and now currently a VP), also working on AI and might be implementing the exact same voice applications for call centers that Mesolitica was building. "Too good of an offer to turn down lah," remarks Khalil, but it created a direct conflict because "this is what I am promoting on the application side of MaLLaM lah. For a call center, vision development, voice applications."

This wasn't just a team member leaving; it was a "huge, ah, wrench and spanner into my whole plan" moment. But what struck me was Khalil's candor. Instead of sugar-coating it, he admitted the strategic shift this forced: "my focus is on upskilling and training. Then, a lot of, even our revenue tahun ni, lebih kepada training. It's not commercial." When your technical co-founder takes your core product idea into a corporate setting, you don't just find a new CTO, you question whether the product-market fit was ever really there.

Mesolitica's migration from Azure to AWS at the end of 2024 wasn't about technology—it was about go-to-market strategy and access. After 17 months of collaboration, they secured a press release from AWS US, not AWS Malaysia. "It just quashed a lot of questions lah," Khalil explained. For non-technical stakeholders, and it is a fact that most decision-makers in Malaysian enterprises are non-technical, seeing US-level validation instantly transformed Mesolitica from "another AI startup" to "AWS-approved partner."

The key insight: In emerging markets, your cloud provider's brand becomes your brand. That AWS US press release was worth more than any technical optimization for opening doors to government contracts and enterprise deals.

The RTM partnership sounded brilliant on paper: access to radio data to train voice models that understand local dialects. "Natural here meaning that even dialects are understood. So that's why the RTM MOU was very important. Because we will have access to the RTM radio data."

But here's the learning moment: The complexities of data governance. RTM retains ownership because the output voice from their presenters belongs to RTM, but the underlying model's learnings can be commercialized. As Khalil mentioned: "Open sourcing the data tu, is RTM okay with that? Tak. Obviously not lah."

This encapsulates the POC trap that Khalil kept circling back to: "Last year, ah, me getting a POC with a big name is a big deal lah. This year, it's not lah. Now, it's like, "eh, if you go with this organization, okay, how fast do we get a contract?"

Everyone wants to do a POC. No one wants to sign a commercial contract. The revenue is "really still not clear" because "it's still an immature technology." And yet, you need these POCs for face time with ministers—a "public relations strategy."

I think all of us working on the building side of technology can empathize with that.

With Husein temporarily unavailable, Khalil activated the "Future is Solo" narrative. At first, it sounded like typical startup pivot-speak. But the more he explained it, the more it revealed a deeper truth about where the AI market is actually heading.

Translation: When you can't sell product, sell the process. When you can't sell AI, sell the ability to build with AI.

This isn't just a public relations strategy: It's become their main revenue driver. "My focus is on upskilling and training. Then, a lot of, even our revenue tahun ni, lebih kepada training. It's not commercial." The "Future is Solo" narrative, with its emphasis on "one human with a thousand agents," resonates because it addresses the real problem that businesses currently have, which is not a shortage of AI tools, but a desperate shortage of people who know how to use them effectively.

You can read more on Khalil's "The Future Is Solo" whitepaper here.

The conversation turned to Japan's aging population where 30% of the populaton is over 60 which is creating healthcare burdens and demand for AI-powered solutions. This demographic reality creates immediate business opportunities for voice AI and automation tools that can serve elderly populations and the caregivers supporting them.

Khalil sees potential for their dialect-understanding voice models to be adapted for Japan's regional variations, though I think the data governance challenges would likely mirror those they faced with RTM.

Here's where Khalil got technical and brutally honest. "For me, I realized around the middle of the year that agentic AI is just RPA plus LLM... Andrew Ng's definition of this is to fully utilize capabilities of LLM to be autonomous and that is still far off."

The distinction between "agentic AI" (planning and execution) and "agentic workflow" (RPA + LLM) is crucial. Most implementations are just the latter, and "it's still too early to be saying yes we can automate everything."

This is the kind of realization that should be shouted from rooftops: We've taken traditional robotic process automation and slapped a language model on top, and suddenly we're calling it autonomous AI. It's not. It's workflow automation with better natural language understanding.

Since we're talking about RPAs and LLMs, I then bring the conversation to generating stuff with AI, and how the agentic workflow view is also affecting our product Kafkai.

"The value is not generating anymore. So our latest pivot is looking at it and figuring out that our clients problem is actually knowing what to generate."

Everyone can generate content with ChatGPT. The moat isn't generation anymore: It's curation, strategy, and understanding what deserves to exist in a world drowning in AI slop. "ChatGPT can't give you this as it has limited memory and limited context. It tries to fake it, but it can't get it right. A problem that most of us have is that we can't differentiate the faking and getting it right."

This is the single most important learning for any content creator or marketer right now. In a world of infinite AI content, authenticity and strategic curation become the ultimate differentiators.

"How do you be authentic? Kena keluar lah" (How do you be authentic? You have to put yourself out there).

Mesolitica maintained their open source narrative even after Husein's departure: Khalil says: "I just offer MaLLaM as open source je." This attracts government clients but raises investor questions about IP value.

The tension is palpable: "investor nak dengar kan you actually have something kat belakang." (The investors want to hear that you have something at the back, like a protected IP or your secret sauce and not just another wrapper or an open-source technology). The AWS press release helped: "It just quashed a lot of questions lah", but the fundamental challenge remains. In a world where Chinese open source models are driving token costs down, making it "tragic for founders" but beneficial for application builders, how do you justify a proprietary moat?

Khalil's answer seems to be: you don't. For him, he pivots to training, to "Future is Solo," to selling the methodology rather than the model.

If you're running a content strategy or bootstrapped startup, Khalil's and Mesolitica's journey offers several concrete takeaways:

1. Brand validation matters more than technical superiority in emerging markets. That AWS US press release was worth more than any technical optimization.
2. POCs are the new pilot projects—but they rarely convert. Budget them as marketing expenses, not revenue pipelines.
3. When product-market fit is unclear, sell the process. The "Future is Solo" narrative turned a crisis into a revenue stream.
4. In Japan, demographics are destiny. A 30% over-60 population creates healthcare automation opportunities that Western markets won't see for a decade.
5. Agentic AI is mostly RPA + LLM. Don't overpay for "autonomous agents" that are just fancy workflows.
6. Content strategy is still content strategy even with AI—but the value is in knowing what to generate, not generating more.
7. Open source attracts government clients but scares investors. Have a clear answer to "what's actually proprietary here?"
8. Authenticity is the final moat. In a world of infinite AI content, your personal brand and real relationships become your only defensible assets.

After nearly two hours of discussing a wide range of topics with Khalil, I realized that my mental bandwidth was at its limit. We decided to say our goodbyes. However, as I reviewed my notes later that night, several lingering uncertainties emerged—questions I suspect many AI founders share:

• How do you quantify the value of AI-generated content beyond conversion metrics when the AI itself is becoming the consumer?
• What's the exact definition and lowest denominator of an "AI builder"? Is it someone who can write prompts, or someone who understands first principles?
• How do you balance open source credibility with commercial IP requirements for investors who want "something at the back"?
• What is the sustainable economic model for serving 1,000-10,000 SME users without burning margins when token costs keep dropping?

These aren't rhetorical questions. They're the difference between building a sustainable business and becoming another cautionary tale in the AI graveyard.

Khalil's story isn't a victory lap—it's a survival manual written in real-time. Mesolitica's pivot from building the definitive Malaysian LLM to selling the "Future is Solo" dream wasn't a strategic masterstroke; it was a forced evolution when market reality tore up their product roadmap. The CTO's departure, the POC purgatory, the investor skepticism—these aren't unique problems. They're the standard curriculum for AI founders in emerging markets who discover that having a better model means nothing if you can't get paid for it.

The brutal lesson? In 2024's AI landscape, your moat isn't your model—it's your ability to teach others to fish while everyone else is still trying to sell overpriced sushi. The AWS press release, the RTM partnership, the open-source credibility; these weren't product features, they were trust-building exercises in a market that doesn't understand the tech but desperately needs the outcomes.

For content marketers, SMEs, and fellow founders, Mesolitica's compressed lifecycle offers a clear takeaway: authenticity and education are the only defensible assets when token costs race to zero and every POC is just a polite no. The "Future is Solo" narrative works because it acknowledges a truth most AI companies won't admit—we're not building autonomous agents; we're building slightly smarter tools that still need human hands to wield them effectively.

The questions that remain, about quantifying value, defining builders, balancing open source with IP, these aren't Mesolitica's alone to answer. They're the industry's growing pains. And if Khalil's right that "the future is solo," then perhaps the most valuable product any AI company can sell right now isn't the AI itself, but the wisdom to navigate these questions without burning through runway.