Tech • AI • Robotics • Game
The LIE Session 3 - My "AI Employee
6/10TL;DR
Artificial intelligence is moving from simple chatbots to tools that can automate work on personal computers and in clinics, but adoption is being slowed by governance, privacy and organizational inertia.
KEY POINTS
Healthcare pilots meet institutional friction
Clinicians are testing ambient listening agents that can capture a consultation and generate a draft note, allowing practitioners to stay focused on the patient rather than documentation. Early enthusiasm is tempered by questions over GDPR, data handling, team-wide adoption and the gap between an individual initiative and an institutionally approved workflow.
Medical use cases depend on specialization
A generic AI summary is often not enough for healthcare. Clinical letters and reports vary sharply by specialty, and a note suitable for internal medicine may not fit orthopedic surgery or another field. That makes customization crucial, whether through prompts, templates or more advanced context systems.
The technology has advanced far beyond simple prompting
The discussion highlighted a progression from basic chatbot use to AI systems connected to external tools, then to software that can act directly on a user’s computer. These systems can create forms, manipulate files, retrieve local information and chain multiple tasks together, turning AI from an assistant into an operational layer for daily work.
Context management is becoming a core capability
A key technical concept is the harness, a layer that gives a language model access to selected files, tools and instructions on a local machine. Its main role is to assemble the right context for the model and expose the right actions. Too little information reduces performance, while too much irrelevant data can also degrade results.
Coding is being reshaped, not eliminated
The emergence of tools such as Open Code, Claude Code, Codex and other agentic coding environments is changing software development practice. Experienced developers increasingly report that they no longer write or even reread every line of code generated for smaller internal tools, though they still stress that software engineering judgment remains important for reliability, security and products intended for broader use.
Automation now reaches everyday media work
One practical example described a local application that automates publication of short videos by removing silences, generating subtitles, scoring vocal delivery and checking legal risk in the spoken content before upload. Tasks that once required domain-specific expertise in audio processing can now be assembled quickly with AI support, significantly lowering the barrier to building custom tools.
Deterministic code is favored for repeatable tasks
Participants drew a clear distinction between deterministic automation and generative AI reasoning. When a workflow is repetitive and standardized, conventional code remains preferable because it produces stable outputs and consumes fewer paid AI resources. Language models are then reserved for the parts of a process that require interpretation, such as analyzing text, tone or legal exposure.
No-code remains an entry point, but with limits
Platforms such as n8n and other no-code tools can help less technical users begin automating workflows. But they were described as less flexible than code-based approaches, especially when users want deep customization, access to local files or integration of several layers of tools and models.
Cost and access still shape experimentation
Running AI through an API rather than a consumer chat interface allows these tools to control software and workflows, but usually introduces usage costs. Even so, free quotas from some providers are making experimentation possible, including in prototypes that embed AI analysis inside private applications rather than relying only on public chat services.
Privacy fears extend well beyond AI vendors
The exchange broadened into concerns over surveillance, health data, insurance and algorithmic influence. Even if more work is moved to local models or self-hosted tools, participants noted that much personal data is still generated by others through messages, photos, social media and connected devices. The debate suggested that AI intensifies long-standing power dynamics rather than creating them from scratch.
CONCLUSION
AI is quickly becoming a practical layer for automation in media, coding and healthcare, but its broader impact will depend as much on institutions, privacy rules and social trust as on technical performance. The central challenge is no longer whether the tools work, but how organizations and individuals choose to govern and integrate them.
Ask a question