Beyond the vibe: Why do we need more SWEs in the age of (Gen) AI

Beyond the vibe: Why do we need more SWEs in the age of (Gen) AI  Dr. Shalini Kurapati  Milan, 15 Oct 2025 Note: I’ve been working with AI since for quite a few years, much before it went mainstream. I think it’s cool, but not always.  This talk is deeply researched, thought-through and validated. Credit where credit is due. Gen AI purposefully not used. All facts and claims backed by data.  What we keep hearing these days Source: The Verge Tech layoffs from profitable Incs Seattle Times and Fast Company reporting Are tech (all) jobs being replaced by AI? Source: The Atlantic and the PC magazine Let’s unpack the promise and hype Won a Nobel or two https://www.nobelprize.org/ Sizing the prize!  https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier#business-value AI = Data + Code + Compute  From Elizabeth Bruce, Microsoft’s slides LLMs driving Gen AI hype - Category of deep learning models trained on massive amount of data - General purpose usually - Type of neural network known as transformers - Transformation input-output through the use of billions of parameters Pre-training and fine-tuning all need data and human feedback.    http://jalammar.github.io/how-gpt3-works-visualizations-animations/ The data challenges We are running out of training data ●No Data No AI  ●Propriety or Closed datasets still key: subject matter expertise needed  ●Synthetic data helpswith caveats  Copyright? What copyright Sources: NY Times, Mogin law LLP and CloudFlare, NPR Security risks Security risks of do-it-now adoption trend arXiv:2501.09431v1 [cs.AI] 16 Jan 2025A Survey on Responsible LLMs: Inherent Risk, Malicious Use, and Mitigation Strategy   IBM Cost of Data Breach report 2025 Data leaks  @Eito_Miya mura on X Poisoning easier than thought (less than 250 docs) ARS tecnica article, Anthropic UK AI Safety institute study, Oct 2025  Vibe coding: From frenzy to friction Source: Business Insider September 2025 Arrrghhh….  Convoluted new career opportunities Source: 404 Media Sep 2025 And there’s risk of atrophy: Delegating thinking https://pursuit.unimelb.edu.au/articles/as-ai-gets-smarter,-are-we-getting-dumber Hard to sustain the current business model Our overreliance on AI tools are powered by “cheap” tokens, are now VC funded. It may not always be the case. The age of AI abundance may soon be over The real fear of the B wordCircular shell financing  -The “Magnificent 7” tech companies are going all in on AI  -2/3rd of S&P 500 index -Data centre investments accelerating  - Circular shell financing The Guardian, Oct 2025 No profits in sight for Open AI et al Reuters, Sep 2025 Energy use Financial times Ansa (Fresh/Drinking) Water use  The Conversation, Sep 2025 ●All Google Gemini median prompts: about 650,000 liters per day. ●All GPT 4o medium prompts: about 8.8 million liters per day. ●All GPT 5 medium prompts: about 97.5 million liters per day  Know this gentleman? Geoffrey Hinton- “Godfather of AI”  - Artificial neural nets - Turing Award, 2018 - Nobel Prize, 2024      Photo by Arthur Petron - CC BY-SA 4.0 AI and Radiologists Source:Creative destructive labs Meme save us Ociubotaru on Threads What do healthcare experts think in 2025 NY times, May 2025 and Mayo Clinic report:     “Despite early predictions that AI would replace radiologists, Mayo Clinic's experience shows the technology is enhancing, not eliminating, radiology jobs by boosting efficiency and accuracy. ” Humans + AI > Humans or AI alone  AI automates System 1 tasks: fast, intuitive pattern work - bug testing, code snippet generation, automating routine and repetitive tasks.  Abstraction, judgment, ethics, teamwork remains human.   ●Humans handle quality assurance, validation, secure architectures, and stakeholder collaboration.  ●Creative reasoning, long-term maintainability, and designing trustworthy systems.   Why LLMs struggle with this simple game NYT Connections uniquely combines linguistic isolation, resistance to intuitive shortcuts, and provides regular updates. AYL Lopez et al and NYT Games Try today’s game on New York Times games What you do is beyond coding NYT Games - Creativity, design, thinking out of the box, innovating: not just pattern matching - Understanding client needs, making sense of feedback, collaboration - Domain expertise, business processes - varied - With the rise of AI, having devs who know what they are doing is crucial to build robust systems - Smaller models with good quality data might be game changers - Demand will increase beyond IT heavy companies. Banking, Finance, healthcare, Agriculture  AI is here to stay, So are you! Future of jobs: 5 year outlook Source: WEF, January 2025 ●1,000 leading global employers ●Representing more than 14 million  ●22 industry clusters ● 55 economies from around the world