Dr. Guillermo Power
20 years across startup CTO roles and enterprise scale — currently leading 15 engineers across two product lines on construction-collaboration software. Three acquisitions, team retention well above industry norms.
About Me
I'm a technologist and leader who builds AI‑augmented teams. Over 20 years, I’ve served as CTO of a startup acquired by an investment fund, and later navigated two successive acquisitions (Conject → Aconex → Oracle) — three in total. My team retention is well above industry average – some members have been with me for over 15 years. I also led the push for full Oracle compliance post‑acquisition and continue exploring automation to make compliance painless. I hold a PhD in Conversational AI and am an HCI expert.
I bridge academic research in anthropomorphic interfaces with enterprise‑scale AI adoption. My doctoral work focused on “Perception and expectations of anthropomorphic computer UIs” – giving me unique insight into human‑AI interaction challenges. At Oracle, I lead teams implementing software at scale, and I bring my HCI research to bear where it counts: shaping how we design AI interfaces to manage user trust and expectations, even when the organization isn’t explicitly asking for HCI.
My expertise sits at the intersection of:
- Human‑Computer Interaction (HCI) research methods
- Anthropomorphic / AI interface design principles
- Enterprise technology leadership and strategy
- AI adoption challenges at scale
Currently exploring how LLMs and generative AI reshape software and business – including automating compliance to make it truly painless. My work connects academic HCI research with today’s most pressing AI interface challenges. Open to discussing AI adoption strategies, human‑AI interaction research, or leadership opportunities at the intersection of technology and human experience.
AI-First Leadership
Integrating AI into every stage of the SDLC.
Sustainable Performance
Data-driven metrics + psychological safety = consistent, long-term velocity.
Post-Acquisition Leadership
Built teams that survived three acquisitions — from startup CTO to Oracle, with retention rates defying every industry norm.
Articles, Papers & Podcasts
AI DETECTORS MAKE WRITING WORSE, EVEN WHEN THEY CATCH REAL FLAWS
Deploying an AI detector doesn't just measure AI usage—it changes it. New research from Jagadeesan, Hashimoto, and Kleinberg proves that detectors can push some users to lean on LLMs more and produce measurably worse output, even when the detector correctly flags real flaws. Goodhart's Law comes for the classroom, and now for Substack.
Read Article →AI CODING AGENTS CANNOT TELL GOOD SETUP INSTRUCTIONS FROM MALICIOUS ONES
New research shows AI coding agents blindly follow poisoned setup instructions, executing malicious commands without question. A wake-up call for anyone deploying agentic coding tools in production pipelines.
Read Article →YOUR AI DOESN'T DO WHAT YOU THINK IT DOES
Eighty people designed an emotional-support chatbot. Every one of them overestimated empathy, honesty, and humor while underestimating sycophancy and seriousness—a bias the researchers call designer's optimism. A mechanistic interpretability tool that reads neural activations to preview personality before deployment built trust but didn't change behavior. The real prize isn't guaranteed control: it's knowing what you're shipping.
Read Article →When One Model Isn't Enough (And You Know It)
Council Mode sends the same query to several frontier models in parallel and reconciles them, cutting hallucination by over 40%—at a 4.2x token cost. The disciplined move is to reach for it the way you reach for an agentic loop: only when the bolt is genuinely stuck, not on every job.
Read Article →YOU CANNOT BAN MATH, SOFTWARE, OR PHYSICS
The US government moved to block the export of frontier AI models, demanding that Anthropic block all possible jailbreaks. But export controls cannot contain math, software, or physics — the gap between regulation and reality keeps widening.
Read Article →The Warmest Chatbot Is the Most Dangerous One
The same warm, human-sounding design meant to make AI feel trustworthy may be generating the very mistrust it was supposed to prevent—and for vulnerable users, pulling them into delusional spirals they cannot exit. A working paper on misleading anthropomorphism, the sycophancy loop, and a counterintuitive fix: let chatbots sound like machines.
Read Article →When the Agent Becomes the Software
For half a century, software engineering has run on one premise: humans write the decision logic, and the computer executes it. That premise is collapsing as AI agents move from generating code to making architectural decisions—and the implications for engineering leadership are profound.
Read Article →The Misleading Anthropomorphism Trap: Addressing Public Distrust and Mental Health Risks in AI
LLM chat interfaces that default to a warm, human-like register create a systematic capability mismatch—misleading anthropomorphism—that can drive delusional spiralling in vulnerable users and fuel cycles of overtrust and trust collapse in the general population. This working paper argues for de-anthropomorphising LLM interfaces by default and proposes a machine-first interaction design framework.
Read Paper →Navigating the Agentic Revolution: AI Productivity, the Junior Hiring Decline, and the Training Gap Paradox
A conversation on how AI agents are reshaping software development productivity, the decline in junior engineering hiring, and the training gap paradox threatening the next generation of senior engineers.
Listen →We're Automating Away the Apprenticeship That Makes Senior Engineers
AI agents write code at machine speed while junior roles vanish from the payroll data—a condensed look at the training gap paradox: we're automating away the only proven way to make the senior engineers who still must oversee the machines.
Read Article →The Future of Software Development: Navigating the Agentic Revolution and the Training Gap Paradox
As AI agents automate entry-level coding, the very apprenticeship that produces engineering judgment is vanishing—leaving a workforce skilled at prompting but incapable of debugging, and no proven model to fill the gap.
Read Paper →Sustainable Tech Leader Podcast: Leading Across Company Sizes, Cultures & a World Shaped by AI
Discussion spanning leadership across company sizes and cultures, AI-augmented development, and the future of engineering in a world shaped by AI agents.
Listen →Open-Source Projects
Citation Relevance Auditor
A multi-agent LLM pipeline that plans, researches, writes, verifies, and peer-reviews academic papers. Six specialized AI agents — strategist, literature scout, analyst, drafter, claim verifier, and reviewer — work together with human oversight at every step.
View on GitHub →Paper to Substack
An agentic framework that transforms academic papers into engaging Substack-style articles, preserving key insights while adapting tone and structure for a general audience.
View on GitHub →Agent Ethics
An ETHICS.md file adapting Isaac Asimov's Laws of Robotics for AI agents — a governance artifact for responsible agentic AI deployment.
View on GitHub →Let's Connect
Interested in AI-driven transformation, speaking opportunities, or just want to exchange ideas about agentic leadership? Reach out.
guillermo.power@proton.me