DESIGNING DETERMINISM IN AI - HOW AI BECOMES PREDICTABLE

Embrace context engineering: manage chaotic AI systems through contextual cues to drive predictable outcomes

Dec 16, 2025 5m 5s

Zbigniew Sobiecki

Expert Insights

Zbigniew Sobiecki offers a fresh perspective on AI-assisted software development, pitching the rise of a new profession, "Context Engineering," that aligns seemingly chaotic AI algorithms with concrete, predictable results. He emphasizes that a shift in mindset from fixed coding structures to creating specifications that manage probabilistic outputs is pivotal in reaping the full benefits of AI.

He focuses on context as the primary guiding factor in selecting operating models for AI development, thereby giving birth to the discipline of context engineering.

Hear Zbigniew elaborate:

On the emergence and role of context engineering in managing and controlling AI development.
How classifications of AI models are driven by context and their response to different conditions, leading to effective model orchestration.
The shift from traditional specifications to "control systems," curating context, and generating prompts to manage probabilistic results in AI development.
The increasing need for specialization with AI agents, tailoring them to specific user journeys, and how this paves the way for autonomous work.
How building systems and environments, creating scaffolding that pushes probabilistic systems towards reliable outcomes, is the new engineering discipline.

We're all, from what I've heard, becoming context engineers one way or the other, trying to figure out which pieces of information to put in what order and how to feed it into the LLM to make sense of it.

Monterail Team Analysis

Here are actionable guidelines to prepare and optimize your software development processes in the era of AI:

Embrace 'Context Engineering': Adapt your old deterministic approaches to managing context and generating prompts, aligning your AI systems to produce desired, predictable outcomes.
Adopt 'Spec-as-Infrastructure' approach: Transition from traditional specification documents to designing control systems that manage probabilistic outputs, effectively becoming the understructure of your AI application.
Master Model Orchestration: Assess and understand different models available with respect to their speed, context, and cost, then select or combine models optimized for your project’s specific needs.
Specify AI Agents’ Roles: Recognizing the inherent diversity within your AI model's abilities, ensure specializations are accurately defined and applied to cater to distinct user journeys or requirements.
Design AI Autonomous Workflow: Equip your AI systems to operate independently yet efficiently towards achieving its objectives; develop a system of checks and balances to ensure desired directions are followed.
Prepare for a New Engineering Discipline: Foster an environment that encourages and recognizes the importance of building systems and environments that direct probabilistic systems towards reliable results.
Stay Ahead of the Curve: Experiment with new ideas and tools in managing AI systems; the field is still young with many discoveries and innovations to be made.