Manifesto

The AI Alignment Manifesto

We believe there must be a better way to address the second-order problems of AI if we’re serious about building a future where increasing the autonomy of AI systems no longer requires sacrificing reliability, efficiency, or responsibility.

I don't understand how you can even think of building an agentic system without the ability to predict the consequences of its actions.¹

Yann LeCun, 2026

A new era of hybrid developer-AI cognitive systems

We have entered the era of AI-native software engineering. For decades, developers have used passive tools to build software. Today, AI agents are becoming active participants in the engineering process itself. This marks the emergence of hybrid developer–AI cognitive systems, where humans and AI agents jointly build and operate software.

Second-order problems

  1. Yet our ability to produce software is advancing faster than our ability to understand its behavior and consequences.

  2. Most agentic systems we build today are disconnected from the reality in which they will operate: only after deployment to production do we discover how they behave under real-world conditions.

  3. The increasing autonomy of AI systems seems to be making us less mindful of our own responsibilities as human beings.

We call for

a future in which increasing the autonomy of AI systems no longer requires sacrificing reliability, efficiency or responsibility.

01Reliability

observing what went wrong→understanding what could go wrong

We believe reliability must evolve from observing what went wrong to understanding what could go wrong by going beyond collecting telemetry or expanding evaluation datasets.

Instead, we must transform production behavior into structured, reusable knowledge that hybrid developer–AI cognitive systems can reason over before releasing changes in the real-world.

As humans, we learn to walk independently by falling multiple times (sometimes painfully), but there's no reason to take a similarly isolated approach to how agentic systems experience failure and learn from their production context.

Every failure should make the next failure less likely to occur because reliability should be a continuously improving capability, informed by everything our systems experience in production.

02Efficiency

more code being generated+more tokens being consumed≠better outcomes

We might be confusing activity with progress because more code being generated and more tokens being consumed do not necessarily translate into better outcomes.

An agentic system that reduces operational costs while degrading the user experience or business outcomes is not necessarily better.

As AI economics become increasingly unpredictable due to the non-deterministic nature of agentic systems operating at scale, efficiency cannot be measured solely by minimizing token costs or maximizing throughput. It requires a continuous value engineering practice.

03Responsibility

AI's ability to actour ability to take responsibility

As AI systems gain greater autonomy, we believe their ability to act must be balanced by our ability to take responsibility for the consequences of their actions.

If technology’s aim is to expand human capabilities without diminishing our ability to understand and direct its consequences, then even with autonomous systems operating within clearly defined boundaries, the humans in the loop who build and/or supervise these systems must retain the understanding and agency required to exercise meaningful control.

At Unomiq, we believe these three pillars are critical to achieving AI Alignment. They form the foundation of our humanistic approach to AI development - a future defined by how reliably, efficiently and responsibly we can deploy autonomous AI.


Early access

For teams that care about building a future where increasing the autonomy of AI systems no longer requires sacrificing reliability, efficiency or responsibility.

By joining the waitlist, you acknowledge that Unomiq will process your email address in accordance with our Privacy Policy.