Programmer Career Guide in the AI Era: Transition Paths and Embodied AI Opportunities

Programmers in the AI era must proactively embrace trends, transition from Web Coding to FDE, and take a rational view of embodied AI and AGI.
Based on a Bilibili career planning livestream, this article outlines transition strategies for programmers in the AI era. The core judgment: developers who don't embrace AI face elimination risk and should prepare in advance rather than passively job-hop. The transition path has three steps: master Web Coding as a foundation, move toward FDE to get closer to business, then use AI to transform life and side hustles. On embodied AI, the speaker sees it as a 100% major trend but high-quality data scarcity as the core bottleneck, with data-side roles as an entry point. On LLMs replacing researchers, human value concentrates in business judgment and creative decisions, while AGI and embodied AI deployment remain distant. Opportunities are no longer limited to tier-1 cities, and tool selection should prioritize cost, availability, and stability.
AI technology is iterating faster than expected, and programmers' career anxiety has shifted from "should I learn AI" to "what to learn, how to learn it, and where to pivot." In a recent career planning livestream on Bilibili, the speaker offered refreshingly blunt takes on campus and experienced-hire transitions, opportunities in embodied AI and FDE roles, and AI deployment scenarios. Drawing on the livestream and industry observations, this article distills several actionable approaches.
"Ordinary Members" Who Don't Embrace AI Are Truly at Risk
When someone in the livestream asked whether "ordinary members" meant those who don't embrace AI, the speaker answered directly: "Members who don't embrace AI—it's this obvious and you still won't embrace it? What are you waiting for?" Harsh as it sounds, it points to a reality: AI tools are shifting from a "bonus" to a "baseline."
The speaker elaborated further: even if your current company doesn't need AI yet, you should prepare in advance—"but when development needs it, you can jump in immediately. That's what preparing for the future means." Waiting until the company falters before job-hopping leaves you in a passive position. This "preparing an exit route" mindset is more pragmatic than simply chasing hot trends.

A detail worth noting: the speaker also left room for nuance. State-owned enterprises and other institutional environments operate at a different pace—if you genuinely don't plan to shake things up, you can treat learning about trends as a side pursuit. But the trend itself is certain; the only variable is whether you're proactive or reactive.
The Transition Roadmap: From Web Coding to FDE
When asked whether "AI" referred to Web Coding and AI application development skills, the speaker laid out three skill tracks:
- Web Coding—a must-have; this is the foundational capability for AI application development.
- FDE (Forward Deployed Engineer)—the direction to consider after embracing business, i.e., an engineering role that goes deep into customer scenarios to implement AI solutions.
- Using AI to transform your whole life—including side hustles and entrepreneurship, extending AI capabilities beyond work.
The order of this path matters: first build a technical foundation, then attach to business, and only then pursue personal monetization. Doing it in reverse—chasing entrepreneurship before accumulating technical depth—easily leads to disappointment.
FDE (Forward Deployed Engineer) was first adopted at scale by data intelligence companies like Palantir. Its core characteristic is that engineers are stationed directly at customer sites, understanding both the tech stack and the customer's business pain points and data environment, responsible for customizing general-purpose platform capabilities into deliverable solutions. Unlike traditional pre-sales engineers or implementation consultants, FDEs typically need full engineering capabilities, able to independently complete the loop from requirements analysis and data integration to model tuning and system deployment. In AI deployment scenarios, the FDE's value is especially prominent: LLM capabilities are general-purpose, but every enterprise's data, processes, and compliance requirements differ. Someone needs to translate "model capability" into "business value." This is why the speaker placed it after Web Coding and before personal entrepreneurship—it's a bridge role between technical capability and business understanding, often demanding more in communication skills and scenario abstraction than pure R&D positions.
Embodied AI: The Data Bottleneck Under a Major Trend
When someone asked about the prospects of "embodied intelligence," the speaker was unequivocal: "This is a 100% major trend." But he immediately poured cold water: the main problem right now is the lack of high-quality data.
"The technical methods and approaches have long been ready, but currently there's no high-quality data." This judgment explains why robot demo videos are stunning but actual products are few—algorithm frameworks aren't lacking; what's lacking is high-quality data collection and annotation in real-world scenarios.
The takeaway for job seekers: companies doing data work in embodied AI are worth considering, but salaries differ from direct R&D roles. Choosing the data side versus the R&D side depends on whether you value getting into the industry or short-term returns.
Embodied AI refers to intelligent systems that integrate perception, decision-making, and action capabilities into physical entities (such as robots and autonomous vehicles), distinct from software agents that operate only in the digital world. Its tech stack typically includes multimodal perception, world models, motion planning, and reinforcement learning control. The industry-wide recognized bottleneck is not the algorithm framework itself but high-quality data: collecting real physical interaction data is extremely costly, requiring teleoperation equipment, motion capture systems, or large-scale real-machine trial and error, and it faces challenges like difficult long-tail scenario coverage and inconsistent annotation standards. By contrast, internet text and image data can be scraped at low cost, but robot action data—grasping, walking, assembly—cannot be obtained through crawlers. This explains why many demo videos look stunning yet are hard to productize—demos often target carefully designed scenarios, while the long-tail distribution of real environments far exceeds training data coverage. As a result, roles in data collection, simulation environment construction, and data annotation have become important entry points into this field.
Will LLMs Replace Researchers?
Some viewers worried that as LLMs get stronger, fewer researchers will be needed in the future. The speaker acknowledged the trend: "LLMs evaluating their own R&D output—that trend already exists." But he judged that most of the time, human value lies in understanding the industry, understanding the business, and then making judgments, decisions, and creative contributions—none of which have been fully replaced.
At the same time, his assessment of AGI was quite sober: "AGI keeps getting hyped, but it's actually still far away—especially embodied AI, which is very, very far away." This "trend is certain, but pace is slower than expected" judgment is more useful as a reference than blind optimism or pessimism.

AGI (Artificial General Intelligence) refers to intelligent systems that match or exceed human-level performance across a wide range of tasks, as opposed to today's "narrow AI" that excels at specific tasks. The industry has no consensus on the definition of AGI itself: some use "can perform most human economically valuable work" as the standard, while others emphasize autonomous learning and cross-domain transfer capabilities. Although current LLMs perform impressively on tasks like language understanding and code generation, they still have clear shortcomings in continual learning, causal reasoning, and physical-world common sense—embodied AI even more so, where the perception-action loop in real environments is far more complex than text generation. Therefore "trend is certain but pace is slower than expected" is a relatively rational judgment: the technical direction is clear, but from demo to large-scale deployment, issues of reliability, cost, and safety still need to be resolved—often on a timescale of years or even a decade.
Location and Tools: Two Overrated Variables
On whether opportunities in the AI era exist only in Beijing, Shanghai, Guangzhou, and Shenzhen, the speaker's view was: in the internet era, job creation was concentrated in tier-1 cities, but in the AI era, central and western provinces also have opportunities. For a one-person company, location barely matters—unless you must recruit from a specific place, in which case you'd need to consider talent cost and quality.

On tool choices: for beginners dabbling, Doubao or Coze are fine; for work scenarios, you can try Work Body; for professional programming, if you have the budget, choose overseas solutions—domestic options like DeepSeekAd and Kimi are sufficient. The core considerations are cost, availability, and stability.
One final reminder came from an outsourced developer after being laid off: "No technical approach today can do without LLMs—not even math." That's an exaggeration, but the direction is right—AI has gone from being a tool to being infrastructure.

One-line Summary
Programmers in the AI era must proactively embrace trends, transition along the Web Coding to FDE path, and take a rational view of the pace of embodied AI and AGI.
Paragraph Summary
Based on a Bilibili career planning livestream, this article distills transition strategies for programmers in the AI era. The core judgment: developers who don't embrace AI face the risk of being left behind and should prepare for the future in advance rather than passively job-hopping. The transition path has three steps: first master Web Coding as a foundation, then move toward FDE (Forward Deployed Engineer) to get closer to business, and finally use AI to transform your life and side hustles. On embodied AI, the speaker believes it's a 100% major trend, but the lack of high-quality data is the current core bottleneck, and data-side roles can serve as an entry point. On LLMs replacing researchers, the article argues that human value concentrates in business judgment and creative decisions, while AGI and embodied AI deployment remain far off. Additionally, opportunities in the AI era are no longer limited to tier-1 cities, and tool selection should center on cost, availability, and stability.
Related articles

Google Releases Gemini 4 Hydrogen: Speed and Code Generation as Core Selling Points
Google releases Gemini 4 Hydrogen, emphasizing speed and first-time code accuracy; agents can pay bills and open Google Ads accounts; Baidu launches model trading platform Ciyuansuo, Alibaba upgrades its 714 flagship, and the 27B-parameter open-source robotics model Prince GE opens for commercial use.

Gemini 4 Hydrogen Long-Term Test: Official 2.5x Speed, 3D Printing Is the Biggest Surprise
Full long-term test of Gemini 4 Hydrogen: priced at $5.5/$7 per million tokens, officially 2.5x faster, stunning 3D-printed STL quality with no supports needed, and only 4% weekly quota used.

Pi Agent's Minimalist Design Philosophy: Why Fewer Features Makes It More Popular
Pi Agent's minimalist-core-plus-plugin design contrasts with the feature-bloated Codex and Claude Code. We analyze its design philosophy and star growth.