Industry is racing to develop and field Collaborative Combat Aircraft (CCA). Airframes are being designed.
Sensors are improving. Artificial Intelligence (AI) technologies are advancing at an extraordinary pace.
The central argument of this paper is that CCA development must treat tactical autonomy as an operational
capability, not simply a technical feature. That requires decomposing combat problems, allocating authority
at the function level, designing for uncertainty and testing behaviours under representative conditions—with
contemporary operators embedded throughout.
But a more fundamental question receives far less attention: Who is developing the autonomy behaviours
that will enable these aircraft to exercise the tactical judgement we expect from a fighter pilot in
representative combat environments—and, one day, exceed it?
A fighter pilot does far more than fly a fighter aircraft. They observe the environment, build situational
understanding, anticipate how the tactical problem is likely to evolve, weigh competing courses of action,
manage risk and direct actions—often across multiple aircraft simultaneously. They do so within the bounds
of commander’s intent, mission objectives, delegated authority and legal constraints, while continuously
balancing tactical opportunity and risk. It's an unforgiving environment and losing is consequential.
Having trained fighter pilots for more than 20 years, I can attest that teaching this Air Domain proficiency
takes years, particularly the cognitive component. Having also taught throughout the transition from 4th to
5th generation aircraft, I've observed that technological advances in aircraft have not fundamentally reduced
the time required to develop this proficiency in humans. If we expect CCA to become genuine force
multipliers in dynamic and uncertain tactical environments, we need a way to identify the cognitive
functions, tactical judgements and decision authorities that underpin effective fighter employment.
The development challenge is to determine which of those functions should be delegated to autonomy, how
they should be represented in system design, and how the resulting behaviours and decision authorities can
be tested and assured.
If a CCA is going to operate as part of a combat team that includes crewed fighters, its autonomy needs to
understand enough about the tactical problem to generate behaviours that are effective in the Air Combat
Domain. By “understand”, I do not mean that CCA must reason like a fighter pilot. I mean that its autonomy
must maintain enough awareness of tactical context, mission intent and constraints to generate appropriate
behaviour.
From a fighter cockpit, I do not particularly care how the autonomy is implemented. I care about what the
CCA actually does.
• Does it position itself where I expect it to be?
• Can it manage sensors without consuming my attention?
• Can it react quickly enough when the threat changes?
• Does it preserve the geometry I need?
• Does it understand when to engage, abort, or when it has reached the limits of its authority?
Most importantly: Does it improve mission effectiveness by making the combat team more lethal and
survivable? That is the operational test.
A fighter pilot rarely performs one discrete task at a time. During an air combat mission, I may simultaneously
be maintaining formation geometry, interpreting the tactical picture, managing sensors, monitoring threats,
tracking fuel and weapons, communicating with other aircraft, assessing adversary intent, and deciding
whether the mission remains aligned with commander's intent. All of those demands compete for finite
attention while the tactical situation continues to change.
That is why asking “Can the autonomy perform this function?” is the wrong starting point. The more useful
question is: “What does the combat team need to do to win this tactical problem?”
One useful bridge between operators and AI engineers is Boyd's well-known OODA Loop. It is not an
autonomy architecture, nor does it fully describe the complexity of tactical cognition. But it provides a useful
operational decomposition framework:
• Observe • Orient • Decide • Act
Used this way, OODA gives operators and engineers a common language for decomposing tactical scenarios.
It helps identify what information must be sensed, what understanding must be developed, what decisions
must be made and what actions may follow. Critically, it also supports decisions about which functions
should be delegated to autonomy, under what conditions and with what level of human oversight.
Many of these individual functions and behaviours are already achievable, at least under defined operating
conditions. The deeper challenge is developing tactical autonomy that can interpret mission objectives,
uncertainty, adversary behaviour, risk, authority and likely future states—and use that understanding to
select tactically appropriate behaviours in contested and denied environments.
Consider a Defensive Counter Air mission. At a high level, mission success may be simple to articulate:
prevent hostile aircraft from successfully attacking the defended asset. But underneath that objective is a
continuous series of tactical functions in which a CCA may need to participate.
The combat team must build and maintain a picture of the threat. It must position assets relative to the
defended asset, adversary aircraft, support aircraft and friendly forces. It must manage sensors and
emissions. It must determine which tracks matter and manage appropriate sensor timelines.
It must preserve favourable geometry. It must react to changes in threat direction, composition and
behaviour. It must make decisions about committing to an intercept, targeting, weapon employment,
disengagement and re-attack. And it must do all of this while managing fuel, weapons, survivability,
communications and time.
A requirements document or functional performance specification can describe a mission. It almost certainly
will not capture the hundreds of tactical judgements, assumptions and behaviours that sit underneath
successful execution. Those details are critical to designing useful CCA autonomy. This is where
contemporary air combat experience matters.
A function may be technically straightforward to automate but provide little operational value. Another may
be technically difficult, but if solved, may dramatically improve mission effectiveness. CCA autonomy should
therefore be prioritised according to the tactical problems it solves for the combat team, rather than
according to what is easiest to automate.
From a fighter pilot's perspective, some of the most valuable autonomy may initially appear relatively
unglamorous. Consider formation management.
If I have to continually command heading, altitude, speed and position to keep a CCA tactically relevant, then
I do not have a collaborative aircraft. I have a remotely managed aircraft creating another demand on my
workload. Useful CCA autonomy should understand the tactical purpose of its positioning well enough to
maintain appropriate geometry within defined constraints.
The same principle applies to sensors. I should not have to continuously tell a CCA where to allocate sensor
timeline, which track to maintain or how to manage routine information collection. Similarly, threat
monitoring is only useful if the CCA can do more than detect that something has changed. It needs to
recognise when that change matters to its assigned role and respond appropriately within its delegated
authority.
One way to prioritise where autonomy may create operational value is to consider three questions:
• How much pilot capacity does this function consume?
• How much contextual tactical judgement does it require?
• What are the consequences if the autonomy gets it wrong?
Functions that consume significant human attention, can be bounded within well-understood tactical
constraints, and have manageable failure consequences may be attractive candidates for early delegation.
Bounded elements of these functions might include:
• Formation and station keeping
• Navigation and deconfliction
• Routine sensor tasking
• Track maintenance and correlation
• Threat monitoring
• Information management
• Fuel, weapons and system-state monitoring
That does not mean these functions are simple. Track correlation, sensor management and threat
assessment can involve uncertainty, conflicting information, incomplete data and adversary deception.
Different elements of the same function may demand very different levels of contextual judgement.
The point is not that entire functions should simply be “automated”. The point is that elements of those
functions may be delegated where the tactical conditions, authority boundaries and failure consequences are
sufficiently understood.
This also introduces the question of reversibility. An autonomous decision that can be rapidly corrected is
different from one whose consequences are immediate and irreversible. The appropriate level of delegation
should reflect that distinction.
At the other end of the spectrum are decisions in which broader tactical context, commander's intent and
consequences beyond the immediate local problem become dominant.
Should the CCA initiate an engagement? Is the emerging threat worth accepting additional risk? Should the
CCA continue prosecuting its assigned task or preserve itself for another purpose?
CCA autonomy may eventually make sophisticated decisions in these areas. But the development question
should not be: “How quickly can we remove humans from the decision?” It should be: “What decision
authority can we usefully delegate, under what conditions, within what constraints, and to what operational
benefit?”
This is perhaps the most important practical point. If fighter pilots and CCA are to fight effectively together,
the autonomy cannot simply transfer work from one interface to another.
Every time the CCA effectively asks “What should I do now?”, its human teammates must stop thinking about
something else. Every time it requires detailed re-tasking, supervision or confirmation, it consumes attention.
Every time it behaves in a way that was not expected, the human teammate must divert attention to
understanding what the CCA believes is happening, why it is acting that way, and whether intervention is
required.
That quickly becomes operationally expensive. The promise of CCA is not realised by adding more aircraft to
the formation if those aircraft demand proportional increases in human control.
As a fighter pilot, I want to communicate intent and the constraints that matter. The CCA should determine
how to execute that task within its delegated authority. That is where autonomy begins to provide combat
mass without generating an equivalent command-and-control burden.
Human-machine relationships may be categorised using terms such as Human-In-The-Loop, Human-On-TheLoop or Human-Out-Of-The-Loop. Those labels can be useful shorthand, but they can also oversimplify the
problem. For CCA to be effective, the aircraft will likely operate under different human-machine relationships
across different functions at the same time.
For example, a CCA could independently:
• Maintain navigation and formation geometry
• Manage routine sensor employment
• Deconflict from other aircraft
• Respond to immediate survivability threats
• Maintain assigned tactical constraints
At the same time, the pilot might retain authority over:
• Major changes to mission tasking
• Changes to tactical priorities
• Commitment into specified threat environments
• Particular targeting decisions
• Weapon employment where human authorisation is required
The important design question is therefore not: “What level of autonomy does the aircraft have?” It is:
“What authority does the CCA possess for this function, under these tactical conditions?”
In training, we progressively give new fighter pilots greater responsibility as they demonstrate competence
and judgement. The authority they exercise on a particular mission, however, also depends on their role, the
tactical circumstances and the commander’s intent they have received. CCA require an equally deliberate
distinction between demonstrated capability and delegated authority.
Those relationships may also change with tactical conditions. The authority appropriate during routine
formation management may be different once a threat is detected. The authority appropriate under reliable
communications may differ from that appropriate during periods of degraded connectivity. The authority
appropriate when the system has high confidence in its tactical picture may be different when information is
incomplete, contradictory or uncertain.
If the CCA’s tactical picture becomes unreliable, the range of actions available to it may need to narrow. The
harder case is when the system’s picture is wrong and it has not recognised the error; testing needs to
examine that too. These transitions must be designed, understood and tested.
Losing communications does not itself grant additional authority. Any action the CCA takes without further
human input must remain within authority already delegated for those circumstances.
A CCA might make a better tactical decision than a fighter pilot. That matters, but it doesn’t settle what we
should authorise it to do. We still need to consider the mission, the risks and who is accountable for the
outcome.
The objective of CCA autonomy design should not be to move every function progressively towards full
autonomy. That is the wrong measure of success. The question is whether the resulting human-machine
team is more lethal and survivable.
• Can crewed assets bring a more potent force to bear without an unsustainable increase in workload?
• Can the formation sense across a greater volume of battlespace?
• Can it better manage risk to mission and friendly forces?
• Can it complicate the adversary's targeting problem?
• Can CCA act at machine speed where human reaction time would otherwise constrain effectiveness?
• Can they continue performing useful tactical functions when communications are degraded?
• And can they do all of this while remaining sufficiently bounded, legible and behaviourally coherent that
their human teammates can confidently anticipate the range of actions they are likely to take?
The goal is not autonomy for its own sake. It is tactical autonomy that generates combat advantage without
imposing an equivalent burden on the human teammates.
The next question is: what happens when the assumptions underpinning those decisions stop holding?
Communications fail. The tactical picture becomes ambiguous. An adversary deliberately attempts to deceive
the system. A behaviour that was appropriate moments ago may now undermine the mission.
From my perspective as a fighter pilot, this is where the real test begins. I need confidence that a CCA can
adapt to a changing tactical problem while remaining within its authority, and that I understand its likely
responses when it reaches its limits.
Mission autonomy that depends on uninterrupted communications and a reliable tactical picture is a flawed
basis for combat employment. When training pilots, we rehearse failures and contingencies because we
expect them to occur. CCA development requires the same discipline.
What happens when the CCA loses contact with its human teammate? When it can no longer establish its
position with sufficient confidence? When on-board and off-board sensor information conflict with each
other?
“Continue the mission” is insufficient guidance. So is “return home”. For example, loss of communications
may trigger a pre-authorised contingency plan, while conflicting sensor inputs may require additional
corroboration or restrict actions that depend on the disputed information.
The appropriate response must account for the mission and the wider team, including situations in which
continuing the assigned task carries less overall risk than stopping it. The requirement is to define which
response is appropriate, what triggers it and what constraints remain in force.
From the cockpit, I need to anticipate that response without having to diagnose the autonomy while
managing the rest of the fight. Graceful degradation should preserve useful capability where possible while
keeping behaviour within understood limits.
A fighter pilot may recognise a tactically attractive opportunity and still be required to let it pass. Mission
priorities, identification requirements, delegated authority or risk to other forces may preclude acting. CCA
need to operate within the same distinctions between what is possible, what contributes to mission success
and what is authorised.
This is why the architecture governing autonomous behaviour matters as much as the sophistication of the
behaviour itself. Commander’s intent and mission objectives should guide what the aircraft is trying to
achieve. Legal requirements, rules of engagement, tactical restrictions and platform limitations must
constrain the actions available to it.
These constraints must operate together. Mission benefit cannot override a prohibition. The test and
evaluation process must assess whether a CCA selects effective actions and rejects attractive actions that fall
outside its delegated boundaries.
A fighter pilot does not necessarily need to predict the exact manoeuvre a CCA will make next. But I do need
confidence about what objective it is pursuing, what constraints it is respecting, what classes of action it may
take, when it will require additional authority, and how it will behave when communications or confidence
degrade. That is the foundation of useful operational trust.
When assessing a fighter pilot’s combat readiness, I look beyond whether they can execute a rehearsed
sequence correctly. I want to observe how they respond when the situation changes, information is
incomplete or the original plan becomes inappropriate. The same principle should apply to evaluating tactical
autonomy behaviours in CCA.
From an operator’s perspective, trust depends on demonstrated performance and an understanding of the
conditions in which the system can be relied upon, including its limitations. A successful demonstration
provides evidence of performance in the conditions demonstrated. Operational trust requires understanding
where that performance holds and where it breaks down.
Testing should expose ambiguity, degraded systems, conflicting demands and adversary deception designed
to challenge the autonomy’s assumptions and manipulate its decision-making.
The human teammate must also be assessed. Can I understand the CCA’s intent? Recognise when
intervention is required? Intervene within the time available? An override function has limited value if its
human teammates cannot recognise the problem and act in time.
Human involvement in setting intent, allocating authority and assessing performance remains essential. Any
requirement for intervention during execution must reflect realistic cognitive and time constraints.
CCA have the potential to increase lethality and survivability, but realising that potential will demand close,
sustained collaboration between Defence and industry. Contemporary operators must work alongside
engineers and test and assurance specialists throughout design, testing and operational employment.
Achieving that requires operators and autonomy engineers to work together at a much deeper level than
simply defining requirements and demonstrating technical performance. Operators need to help define the
tactical problems, behaviours, authorities, constraints, failure conditions and measures of effectiveness.
Engineers need to determine how autonomy can best satisfy them. And both need to continually test those
behaviours against tactically representative scenarios and an adversary actively trying to break the
assumptions on which the autonomy was built.
As the tactical environment changes and adversaries adapt, CCA autonomy will need to evolve. Each update
will require renewed scrutiny of how the aircraft behaves and contributes to the combat team. Operational
expertise must therefore remain embedded throughout development and sustainment.
The measure of success is a combat team that performs more effectively under pressure: decision authority
is clearly assigned, behaviour remains within defined bounds, and capabilities and limitations are understood
by those who fight alongside the aircraft.
That is the standard I would expect of any wingman. Delivering it for CCA will require operational judgement
to remain at the heart of the capability.
Disclaimer: The views and opinions expressed in this article are solely those of the author
