What an AI C-Suite Can Actually Do

HOVR, a Canadian mobility company founded by Harrison Amit, publicly names four AI executives. Mia covers operations, Tia technology, Sia strategy and marketing, and Ria regulatory and administrative work. Harrison has also described a much larger pool of agents that can be created for a job and retired when they drift.

The striking part is not the titles. It is the possibility that software can hold context, coordinate recurring work, use company tools, check routine output, and surface exceptions without a leader relaying every step. HOVR still has human executives. The credible model is software beneath accountable leaders, not software replacing them.

Start with one loop, not an AI org chart

Most companies do not need software wearing executive titles in operations, finance, or marketing. They need relief from one repeated management problem: the same context gets relayed, status gets chased, routine work gets checked, and exceptions get routed by hand.

A useful AI layer can carry part of that loop. It can remember the operating rules, prepare the next step, call a bounded tool, record what happened, and ask for help when the normal path breaks. It should not decide who gets hired, move money, make a legal commitment, or speak for the company simply because someone gave it an executive title.

Think of it as a control room, not a boardroom. A control room gathers signals, routes work, and flags exceptions. The analogy stops at judgment and accountability. Software can prepare a decision. A person must still understand and own its consequences.

HOVR shows both the opportunity and the limit

The strongest current account comes from Harrison's May 2026 ChannelNEXT presentation. An independent event report says HOVR automated dispatch, onboarding, marketing, and data-structuring workflows. Harrison described Mia as an assistant that evolved into an AI COO able to query data and support decisions. He also said agents are retired when they drift or produce misinformation. Read the ChannelNEXT case report.

HOVR's public site lists the four AI roles beside a human leadership team. It also claims a 75 percent lower operating cost. That overlap is the useful clue. An AI role can organize operational work while human leaders retain accountable authority. Review HOVR's public site.

PureBrain appears to supply part of the system. Its product description emphasizes persistent memory, tool access, and coordinated agents. In a vendor-hosted testimonial, Harrison said Mia produced useful results about eight times out of ten during beta. That is a self-reported result, but it makes one point clear: useful AI work still needs checks. See the PureBrain platform and testimonial.

The historical record adds an important counterweight. HOVR's December 2024 offering document listed a conventional human management team and treated an internal AI system for operations and logistics as a future milestone. HOVR may have built a substantial agent layer after that filing. The record does not support a claim that software held corporate authority from day one.

The evidence supports a medium-confidence conclusion: HOVR appears to run a substantive, human-governed AI operating layer. Public sources support recurring work across executive functions. They do not establish replacement of corporate authority or final human judgment.

The control room separates work from authority

A useful model has three layers. Human leaders set policy, priorities, risk limits, and approval rules. Functional AI roles hold context and coordinate recurring work. Specialist workflows handle narrower tasks beneath them.

Work should move back up with evidence. The system surfaces exceptions and records what happened. Requests for consequential action stop at a visible human boundary.

The boundary follows consequence. External communication, money movement, legal commitments, employment decisions, policy changes, safety, and irreversible actions need a person who understands the context and bears the result. Research, reconciliation, drafting, testing, and status work can sit deeper in the system.

This division is consistent with the NIST AI Risk Management Framework. It asks organizations to distinguish human and AI responsibilities, define oversight, and keep leadership responsible for AI risk. It does not prescribe an AI C-suite. It does reinforce the human boundary that makes one governable.

That is governance in plain language: decide what the system may do, what evidence it must return, when it must stop, and who has the final say.

What this looks like on Monday

Take a weekly operating review. Today, someone may pull numbers from several systems, chase owners for updates, compare the plan with what happened, and assemble a report for the leadership meeting. The work is important, but much of the coordination repeats.

A bounded AI layer could collect only the approved records, flag missing or conflicting information, ask the responsible owner for an update, and prepare a short decision packet. It could show what changed, what remains uncertain, and which exceptions need attention. Each step would leave a record so the reviewer can see where the answer came from.

The system would not quietly change the budget, reset a target, or send instructions in a leader's name. A person would review the evidence, decide what matters, and approve any action with material consequences. That is the minimum useful outcome: less time gathering and relaying context, without hiding the decision or moving accountability into software.

Five tests separate useful systems from executive theatre

A title proves almost nothing. A useful system should pass five practical tests:

1. Recurring responsibility. It owns a defined piece of work, not a vague executive persona.
2. Retained context. It can use approved policies, records, and prior decisions without starting from zero.
3. Bounded access. Its data and tools match the job, and prohibited actions are explicit.
4. Visible checks. Work leaves evidence, exceptions surface, and false completion can be caught.
5. Human ownership. A named person approves consequential action and answers for the outcome.

These tests matter more than whether the system is called an adviser, board observer, executive, or team. The public cases range from symbolic titles to real operating infrastructure.

International Holding Company's Aiden Insight is explicitly a non-voting board observer. It analyzes company data and recommends actions, while the board decides what to adopt. NetDragon calls Tang Yu a virtual rotating CEO, but its formal reporting describes pre-execution, evaluation, and decision support. Dictador's humanoid Mika sits at the theatrical end, where the title is more visible than evidence of ordinary operating control.

Small-company accounts add the missing mechanics. SmartQix describes persistent sessions, task IDs, inspection, and human approval for customer-facing work. Jackson Yew describes role instructions, daily checklists, shared tickets, budgets, and logs. Aiprosol documents an approval queue along with hallucinations, tone drift, and brittle handoffs. These are founder or vendor accounts, not independent audits. Their value is showing the controls and failure modes that polished titles conceal.

The strongest evidence comes from bounded work

Claims about AI running a company are hard to test. A bounded delivery loop gives leaders clearer measures: elapsed time, acceptance, exceptions, cost, quality, and human review.

A 2026 Itaú Unibanco case-study preprint reports that one staff engineer, supported by four specialized AI roles, delivered a project scoped for four people in half the planned time. The paper reports 90 percent first-review acceptance of AI-generated code and full integration-test passage. It is one proprietary case, not a universal productivity law. The experienced engineer still held the knowledge, validated the work, and owned integration.

Klarna shows both the upside and the correction at function scale. In May 2024 the company said its AI assistant had engaged with more than four million customers and would produce US$40 million in annualized savings. Later reporting said the company had over-indexed on cost reduction and was restoring human capacity. The repeated claim that AI simply replaced 700 jobs turns a changing service model into a false one-cause story. See Klarna's results release and the Reuters account of its correction.

HOVR's roughly 250 agents, SmartQix's 19, Jackson Yew's 12, and Aiprosol's 10 are not comparable headcounts. One system may count a durable role. Another may count every temporary process or specialist prompt. Count responsibilities and outcomes instead.

A 2026 preprint called CEO-Bench shows why that distinction matters. Models produced structurally valid plans in simulated executive decisions but struggled with conflicting advice, ambiguity, and historical context. It is a simulation, not proof that deployed systems fail the same way. It is evidence that a plausible org chart does not establish executive judgment.

Your first AI executive role should be a workflow

Choose a loop where coordination consumes meaningful time, the inputs are available, the output can be checked, and mistakes can be contained. A weekly operating review, customer follow-up queue, recurring reconciliation, or compliance-preparation process is a better starting point than a digital executive persona.

1. Name the loop and owner. State what repeats, why it matters, and who answers for the result.
2. Define the input and useful output. Avoid a broad instruction such as “run operations.”
3. Limit data and tools. Give the system only what this loop requires.
4. Set checks and exceptions. Decide what evidence proves completion and when the system must stop.
5. Measure review burden. Track whether quality, cycle time, cost, or risk improves after human checking.

Do not proceed when the source data is unreliable, nobody owns the decision, the output cannot be checked, or an error could cause harm before a person can intervene. The three-zone decision framework can help place that boundary.

If the loop reliably prepares the right evidence, catches routine exceptions, and reduces the time a person spends relaying context, the layer is creating capacity. If a leader must recheck everything from first principles, it has moved work rather than removed it.

Map one loop before you choose the roles or tools. Write down its owner, inputs, output, permitted actions, checks, exceptions, and approval point. The AI audit logs guide shows how to make the work reconstructable once the loop begins running.

Sources and limits

Sources were checked on July 31, 2026. Company and founder claims are attributed. Formal disclosures and research papers receive more weight than titles, testimonials, or promotional descriptions. Some pages can change after review.

• HOVR and Harrison AmitChannelNEXT report, HOVR site, PureBrain, and December 2024 offering document.
• International Holding Company / Aiden InsightInitial non-voting board-observer disclosure, Q3 recommendations, and workflow expansion.
• NetDragon / Tang YuCompany role description, 2023 annual report, and 2024 annual report.
• Coordinated small-company teamsSmartQix, Jackson Yew, Paperclip, and Aiprosol.
• Measured compression and capability limitsItaú one-person-squad preprint, Klarna results, Reuters on Klarna's correction, and CEO-Bench preprint.
• Theatrical contrastReuters image record and Dictador-supplied announcement.
• Governance boundaryNIST AI Risk Management Framework Core.

Harrison Amit's LinkedIn wording was not used. The profile was inaccessible during the underlying research, and this review did not use an alternate route to bypass that restriction. HOVR's internal prompts, logs, permissions, override rate, and audited productivity or cost data remain unavailable.