A critical discussion is unfolding across various sectors regarding the optimal integration of artificial intelligence with human expertise, underscoring the need for a balance between technological adoption and human oversight.
GitHub CEO Thomas Dohmke recently emphasized the enduring importance of core manual coding skills for software developers, even as AI tools become ubiquitous. Speaking on “The MAD Podcast with Matt Turck,” Dohmke articulated a vision where AI serves as a powerful co-pilot, generating code and submitting pull requests, but developers retain the crucial ability to scrutinize and modify this AI-generated output. This hands-on capability, he argued, is vital to circumvent potential productivity bottlenecks. Dohmke specifically warned against an over-reliance on automated agents, noting that attempting to explain complex changes through natural language prompts can be significantly less efficient than direct code manipulation. “The worst alternative is trying to figure out how to provide feedback or prompt to describe in natural language what I already know how to do in programming language,” Dohmke stated, highlighting the invaluable efficiency of direct programming knowledge.
This perspective aligns with a growing industry consensus favoring a hybrid approach to AI in coding. Research from Deloitte supports this, indicating that developers leverage AI tools for specific, often repetitive tasks like boilerplate code generation, thereby enhancing daily productivity by an estimated 10-20 minutes. This “trust and verify” strategy is gaining traction, especially given that approximately half of all AI-generated code is reported to contain partial errors, necessitating human review and correction. Google, a pioneer in AI integration, confirms this trend, with over 25% of its codebase now AI-generated, yet still subject to rigorous human oversight and refinement. This nuanced integration suggests that the most successful AI implementations in development will augment, rather than outright replace, human programming expertise.
The role of the developer is consequently evolving. Rather than facing obsolescence, programmers are transforming into orchestrators of sophisticated AI-assisted development workflows. Industry experts foresee a bifurcation of developer roles into product engineers, who primarily leverage AI for rapid code generation, and high-coding architects, who are responsible for ensuring the overarching quality, security, and maintainability of software systems. This shift demands new competencies, including strategic problem-solving, effective communication with AI systems, and high-level architectural design, moving beyond the manual writing of every line of code. The persistent shortage of software engineers, coupled with AI tools’ particular benefit to junior developers, implies that AI will play a pivotal role in bridging talent gaps while simultaneously creating advanced opportunities for seasoned programmers. This mirrors historical patterns in software development where new technologies and abstractions have consistently reshaped work processes without eliminating the fundamental need for human ingenuity.
However, the burgeoning trend of “vibe coding”—a term coined by OpenAI cofounder Andrej Karpathy to describe an excessive reliance on AI-generated code—presents a dichotomy: rapid prototyping versus potential quality degradation. While AI tools facilitate agile development and accelerated iteration, they also introduce significant concerns regarding code quality, the potential for security vulnerabilities, and long-term maintainability. Real-world incidents have already demonstrated the perils of deploying unverified AI-generated code, particularly concerning latent security flaws. This paradox is especially pertinent for startups, where non-technical founders might be tempted to build complex systems primarily with AI-generated code, potentially accumulating technical debt that could impede future scalability and growth. Established technology companies, in contrast, have demonstrated that effective AI integration mandates a careful equilibrium between automation and stringent quality assurance protocols, a lesson that smaller organizations are increasingly learning.
Beyond software development, the strategic investment landscape is also being reshaped by AI. Meta Platforms’ recent US$14.3 billion investment in Scale AI, securing a 49% stake, highlights the growing importance of high-quality training data for artificial intelligence development. Scale AI, a key provider of training data for AI companies, will see its CEO, Alexandr Wang, join Meta’s newly formed “superintelligence” unit. While Meta emphasizes Scale AI’s continued operational independence, the substantial stake grants Meta significant influence. This deal, however, has had immediate repercussions, with Scale AI’s major clients—including Google, Microsoft, and OpenAI—reportedly beginning to withdraw their business. Their concern stems from the potential for Meta, a direct competitor in the advanced AI race, to gain indirect access to their proprietary datasets and pre-release product information through Scale’s data labeling contracts. This shift underscores the critical role of neutrality in the competitive AI data market, prompting rivals like Turing, Labelbox, Handshake, and Mercor to experience a surge in demand. Some AI startups are even contemplating bringing data-labeling operations in-house to enhance security and reduce vendor dependency, while others may follow Meta’s lead in acquiring data providers to secure strategic leverage. For Meta, this investment is a calculated move to bolster its position in the AI race, address internal frustrations over its pace of AI development, and secure a robust pipeline of high-quality training data and experienced AI talent.
In a parallel but distinct market, the automotive industry in Indonesia is also witnessing significant strategic investments and evolving competitive dynamics. Toyota, a long-standing dominant force, has invested US$120 million for a 40% stake in Astra Digital Mobil, thereby gaining partial ownership of prominent used-car platforms OLX and OLXmobbi. This move comes amidst a challenging period for new-car sales in Indonesia, which saw a 13.9% drop in 2024 compared to 2023. In contrast, the used-vehicle market is thriving, with 1.8 million used vehicles sold last year versus approximately 800,000 new cars. Used vehicles offer a more flexible and potentially more profitable path to market expansion, with gross margins ranging from 5% to 15% compared to 2% to 5% for new vehicles. Toyota’s strategy echoes its ventures in other markets, where it operates online and dealership-based certified used-car sales. The alliance with Astra, which already operates BMW Astra Used Car, aims to leverage Toyota’s unparalleled nationwide reach, with over 360 dealers across 143 Indonesian cities and a 33.4% market share in new car sales in 2024. This extensive offline presence is seen as crucial for building consumer trust in big-ticket items like used cars.
Despite the formidable entry of Toyota and Astra, existing players like Carro and Carsome believe the Indonesian used-car market is too fragmented for a winner-takes-all scenario. These startups, which initially focused on online sales, have adopted an online-to-offline model with showrooms and inspection points, though their physical footprint is smaller than Astra’s over 30 dealerships. A key differentiator for Carro and Carsome is their rigorous certification processes, based on thorough inspections and quality standards, which build trust. Furthermore, financing is a critical component, with around 70% of car purchases in Indonesia made through credit in 2022. While startups like Carro, Carsome, and Moladin offer embedded financing, Astra also provides competitive car loans through Toyota Astra Finance, potentially at lower interest rates for certified used cars. Ultimately, the competition hinges on profitability and resourcefulness. Both Carro and Carsome have achieved positive EBITDA, with Carro reporting approximately S$40 million (US$31 million) for FY2025 and Carsome posting US$4.3 million for Q1 2025, with an adjusted EBITDA of US$10.5 million for 2024. Industry experts suggest that smaller, more agile players can fine-tune their cost structures more effectively, enabling them to achieve profitability with smaller sales volumes. This dynamic landscape indicates that while large players bring significant resources, market adaptability, operational efficiency, and sustained customer trust will be key to success in both the evolving tech and automotive sectors.
The broader AI landscape continues to evolve rapidly, with AI agents being a significant area of focus. Companies like Singapore-based SapiensAI are innovating by developing their own “code-like” language for internal communication between agents within their Agnes AI application, aiming for faster execution, higher accuracy, and reduced token usage (40% to 70% fewer tokens). This contrasts with Manus AI, which, while excelling in research, reportedly struggles with basic tasks like CAPTCHAs, highlighting the current limitations of AI agents in navigating human-centric digital infrastructure. The ongoing Asia Economic Summit 2025, featuring leaders shaping Asia’s economic future, underscores the region’s commitment to understanding and integrating these technological shifts. As Tech in Asia continues to track investment trends, generating lists of active investors in various sectors, including fintech and startups in Japan, the focus remains on identifying the firms that are not just writing checks but actively shaping the future of technology and industries.
