The argument
A boutique should earn its headcount.
My view is deliberately provocative: a boutique advisory firm with more than 15 permanent people should have to explain why. If its growth mainly adds research layers, presentation production and internal coordination, I see bloat. AI gives us a reason to redesign that work before hiring around it.
Fifteen is my challenge to the operating model, not a scientific cutoff. A specialist firm can justify a larger core through scarce expertise, client continuity or complex delivery. But the explanation should be the value each role creates. Clients should be buying judgment, ownership and execution—not funding an organisation chart.
For Barry Advisory, the direction is a founder-led senior core, with the option to add two or three senior partners over time, supported by AI and a broad specialist network. That is a direction for growth, not a claim that those partners are already in place.
The evidence changes the team-design question.
The Harvard/BCG experiment with 758 consultants found that, on tasks within the tested AI system’s capabilities, AI users completed work over 25% faster, with higher assessed quality. The authors also showed that capability varies by task. This supports redesigning specific workflows; it does not establish how many people a firm should employ. [1]
The published P&G experiment is especially relevant. Individuals using AI matched the performance of two-person teams without it on product-innovation work. Human judgment remained valuable when selecting ideas. That is evidence that AI can supply some benefits of collaboration in a defined setting—not that one person can replace any team. [2]
My inference is that a senior operator can now carry more of a mandate directly. Research, synthesis, scenario preparation and initial analysis can move closer to the person making the recommendation. Fewer handoffs can mean more context retained and faster decisions. The test is accepted work and client outcomes, including review time and rework.
There is counterevidence to easy productivity claims. METR’s early-2025 trial found experienced open-source developers took 19% longer with the tools tested. Its February 2026 update suggested gains had improved, but selection effects made the newer estimate unreliable. The lesson is to measure the actual workflow and current tools, rather than assume a universal productivity multiplier. [4] [5]
A small core can mobilise a large delivery team.
Permanent advisory headcount and project execution capacity answer different questions. One pays for an enduring institution. The other assembles the skills a particular mandate needs, for as long as it needs them.
Barry Advisory can assemble project teams of 35–50 people for larger mandates, with the right specialist and partner setup. The size depends on scope, budget, timing and agreed mobilisation. This is project delivery capacity through the network, not a claim of 35–50 permanent employees or an immediately available bench.
A commercial turnaround may need a senior advisor and a few functional specialists. A venture launch or enterprise rollout may need engineering, product, design, data, operations and programme leadership working together. Both can sit behind a small accountable core.
For CEOs, VC firms, LPs and family offices supporting portfolio companies, the useful questions are practical: who owns the outcome, which specialists are required, what remains with the client, and when does the team reduce or hand over? Our engagement model makes those decisions explicit.
From the solo-founder unicorn to the zero-employee firm.
Sam Altman’s discussion with Alexis Ohanian helped popularise the possibility of a one-person billion-dollar company; Ohanian subsequently discussed that prospect in his own published commentary. It was a prediction about what AI might enable, not evidence that every founder can achieve it. [3]
I believe a solo founder reaching a billion-dollar valuation is a credible possibility. The more useful opportunity is broader: a founder building a durable, profitable business with far less fixed headcount than the old operating model required. Valuation, revenue and cash generation remain different measures.
I also expect firms with no salaried employees to become more feasible as agents improve. Here, “zero employees” means a founder or owners coordinating software, AI and contracted specialists. It does not mean zero human work, zero suppliers or zero accountability. Fully autonomous, ownerless companies are a different and much more speculative proposition.
The constraints still matter: customer trust, distribution, domain expertise, working capital, security and the ability to recover from mistakes. AI can lower the effort needed to create and operate a product. It does not give every product a market.
The next practical step: a private AI operating layer.
My proposed “AI box” is a client-controlled system that helps leadership understand the business already inside its ERP, source code and accounting records. It could be deployed on premises or in an approved private environment. This is an operating concept to scope and validate, not a claim that Barry Advisory sells a finished plug-and-play appliance.
- Inspect an agreed scope
- Use authorised, read-only connections to selected records and repositories. Preserve source-system permissions, exclude secrets and unnecessary personal data, and show where each finding came from.
- Turn evidence into decisions
- Surface possible revenue leakage, manual reconciliation, duplicated software or process bottlenecks. Treat findings as hypotheses until an operating owner validates them against the source.
- Choose and route models deliberately
- Compare models on the client’s tasks, languages, cost and reliability. Keep restricted workloads within the approved boundary. An unavailable local model must not silently trigger an external API fallback.
- Separate analysis from action
- Require named approval for payments, accounting entries, code releases and operational changes. Record decisions, test recovery and retain a human escalation path.
Sovereignty requires control over access, operations and data flows as well as hosting location; IBM’s architecture perspective makes that distinction explicit. [6] Our sovereign AI and privacy approach connects model choice and deployment to applicable local requirements and the client’s legal and security owners.
The economic test is cost per accepted outcome, including integration, compute, review, support and maintenance. Start with one useful workflow, prove its value and expand only when the evidence warrants it.
C-level as a service, with someone accountable.
The executive layer can change too. A business may need experienced product, technology, growth or transformation leadership without needing every role full-time. AI can prepare operating reviews, maintain decision records and flag exceptions; a named senior leader owns the priorities and acts within agreed authority.
That is our interpretation of C-level as a service: fractional executive responsibility supported by AI and specialist delivery. The mandate defines decision rights, availability, escalation, measures of success and handover. A model’s output does not become a board decision simply because it sounds confident.
This creates a useful option for investors and family offices: put experienced operating leadership into a portfolio company for a defined challenge, with the ability to scale execution around it. It also creates a route from a diagnosed problem to a venture.
Where an internal tool solves a repeatable problem, a spin-off may be worth exploring. First establish customer demand, intellectual-property rights, data permissions, unit economics and an accountable founder. The AI system can help identify and test the opportunity; owners decide whether to invest and separate the business.
My ambition for a modern boutique is simple: keep the permanent core small, use AI where it earns its place, and bring the right people to the work. Scale should follow the mandate. Clients should be able to see exactly what that scale contributes.
Explore venture and project delivery, capital and ecosystem advisory, or discuss an operating mandate.
References & reading
The research informs this opinion; it does not validate a 15-person threshold, a guaranteed valuation or a timetable for zero-employee firms. Sources checked on 19 October 2026.
- Harvard Business School: Navigating the Jagged Technological Frontier (2023). Field experiment with BCG consultants.
- Dell’Acqua and coauthors: The Cybernetic Teammate (Organization Science, published online 12 June 2026). P&G experiment on AI and teamwork.
- Alexis Ohanian’s commentary following his interview with Sam Altman (2024). A primary source for the solo-founder unicorn discussion; an opinion, not a productivity study.
- METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (10 July 2025).
- METR: We are Changing our Developer Productivity Experiment Design (24 February 2026). Read alongside the earlier result.
- IBM: Building AI-ready sovereign platforms (2 June 2026). A vendor’s technical architecture perspective, not independent evidence of business returns.