1. Intent & Reality

Every autonomous task begins with two things that sound simple but are easy to confuse: what we want to happen and what is actually true right now.

AG Arch keeps those two things separate.

The first is intent. Intent describes the result someone is trying to achieve. It gives the work direction: what should change, what problem should be solved, or what outcome should be reached.

The second is current reality. This is the state of the relevant systems and environment before the work begins. It describes what exists now, not what we expect to exist, what existed yesterday, or what an agent assumes to be true.

That distinction matters because autonomous work can only be governed reliably when it starts from reality rather than assumption.

Intent gives the work a direction

An autonomous agent needs more than a general instruction such as “fix this”, “update the system”, or “make it work”.

Those instructions may be enough to begin reasoning, but they are not enough to govern an outcome.

The intent has to establish what the work is trying to accomplish. It does not necessarily describe every action the agent should take. In fact, one reason to use autonomous systems is to allow them to decide how to perform work inside defined boundaries.

AG Arch therefore treats intent primarily as a statement of purpose, not as a detailed sequence of commands.

For example:

Reduce the risk caused by an incorrect production configuration.

That describes the purpose of the work. It does not yet say which file should change, which tool should be used, or exactly how the change should be made.

Those decisions can come later.

Reality tells us where the work actually begins

Intent alone is not enough.

Before deciding what should happen next, the system needs an accurate view of the situation it is acting on.

Suppose an agent has been asked to correct a configuration. The configuration may already have changed since the request was written. Another process may have modified it. The service may be running a different version than expected. A dependency may be unavailable. A previous attempt may have partially succeeded.

If the agent begins from an outdated assumption, even a technically correct action can produce the wrong result.

AG Arch therefore grounds execution in current truth.

Current truth means the authoritative state that can be established from the systems involved in the work.

Depending on the kind of work, that might mean the current state of a repository, an application, a database, an infrastructure service, a business system, or another external source of record.

AG Arch does not need to become the owner of that information. The original systems can remain authoritative. The important point is that autonomous work begins from their actual state rather than from an unverified internal assumption.

Why intent and reality must remain separate

A common failure in automation is to treat the requested outcome as if it already described reality.

For example:

“Change setting A from `off` to `on`.”

This instruction appears straightforward. But before acting, several questions remain.

Is setting A really `off` now?

Does setting A still exist?

Has another change already enabled it?

Is the environment the one the request refers to?

Would changing it now affect something that was not true when the request was created?

The requested change represents intent. The answers to those questions represent reality.

AG Arch needs both.

If reality already matches the intended result, execution may not need to make the expected change at all.

If reality differs from what the original request assumed, the plan may need to change before any action is taken.

That is why the lifecycle does not begin with “execute”.

It begins with understanding where we want to go and where we actually are.

The agent does not get to invent the starting state

Autonomous agents are useful because they can reason with incomplete information. But uncertainty should not silently become fact.

An agent can propose an interpretation, investigate a situation, or request additional information. What it should not do is turn an assumption into authoritative current truth simply because that assumption makes the task easier to continue.

This separation is important throughout AG Arch:

reasoning can propose what may be true; evidence establishes what is treated as true for governed execution.

That principle becomes even more important later in the lifecycle when actions must be observed and outcomes verified.

What happens when reality changes?

Current truth is not necessarily permanent.

Real systems continue to change while autonomous work is happening. Other people, services, agents, or external events may alter the environment.

AG Arch therefore treats current reality as something that can need to be refreshed rather than as a snapshot that remains valid forever.

If an important assumption becomes invalid before execution, the system should not blindly continue with the original plan.

The new state becomes part of the decision about what happens next.

This is one of the reasons AG Arch includes recovery and replanning later in the lifecycle: autonomous work must be able to respond when reality stops matching the conditions under which a plan was created.

Example — changing a system configuration

Imagine an autonomous agent has been asked to correct a configuration used by a running service.

Intent: The service should use the approved configuration and operate correctly after the change.

Before planning the work, AG Arch establishes the relevant current reality:

• which configuration is currently active;
• which version of the service is running;
• whether the requested change has already been made;
• whether another change is currently in progress;
• what authoritative system reflects the actual state.

Suppose the original request assumes that the old configuration is still active, but the authoritative system shows that somebody already changed it.

The agent should not repeat the planned action simply because the request says to do so.

The correct starting point is the reality that exists now.

This may mean that the work is already partly complete, that a different action is required, or that no change should be made at all.

What this stage gives to the next one

At the end of Intent & Reality, AG Arch should have two things clearly separated:

What are we trying to achieve?

and

What is true now?

Only then can the next question be answered properly:

What would count as success?

That is the purpose of the next lifecycle stage — 2. Define Success.