The Future of Expert Advice
Introduction
Large language models (LLMs) and apps like ChatGPT are changing how we generate content and seek advice. All industries know change is coming. The consulting industry, which traditionally thrives on human expertise delivered person-to-person, is no exception. However, the challenge is ensuring that AI-generated advice is accurate, trustworthy, and up-to-date. While generative AI can be astonishingly fluent and knowledgeable in general, it has well-known limitations that must be addressed before it can truly augment or replace human consultants in a reliable way.
The Challenge: Hallucinations and Outdated Knowledge
It’s well publicised that LLMs sometimes hallucinate. They produce information that sounds plausible but is false or fabricated. For example, when AI coding assistants confidently generate references to code libraries that no longer exist or are heavily outdated. They also lack up-to-date knowledge beyond their training cutoff. In fast-moving fields, an LLM will often confidently offer outdated advice. These issues – making up facts and using out-of-date knowledge – present serious risks if one blindly trusts the AI’s output.
Additionally, because the training data comes from mixed-quality sources, a generic LLM’s advice can be unpredictable – excellent in some cases and misleading in others. The model has no built-in way to distinguish a top-tier report from a dubious forum post. For a business leader, this inconsistency makes it risky to trust an unaugmented AI advisor.
All of this undermines trust in AI. Clearly, we need ways to ground AI outputs in verified, current, domain-specific knowledge, and that’s where emerging solutions come in.
A New Paradigm for AI Knowledge Integration
Two new open standards are rapidly emerging, creating new ways to solve these challenges.
One of the most promising solutions is the Model Context Protocol (MCP). In simple terms, MCP is an open standard that allows AI assistants to securely connect to external data sources, tools, and knowledge bases in a consistent way. Many people are describing MCP as “like a USB-C port” for AI applications – a universal interface for plugging in the data and capabilities an AI needs. Just as a laptop can connect to many peripherals through a standard port, an AI agent can connect to many MCP servers that provide specialised information or functions.
With MCP, organisations can host servers that expose certain data or skills to an AI. The AI (acting as an MCP client) can discover what a server offers and then query it for content or perform actions through it. MCP Servers are rapidly appearing as developers and organisations explore the possibilities. The quality varies greatly, and subtle differences in desktop (stdio) and server (SSE) variants is already causing confusion demonstrating that there is some maturing required in this space.
So far, much of the hype around MCP has been about using it to give AI agents tool-driven capabilities (like booking a trip via a travel API or managing emails via an email API). These action-oriented use cases are exciting but face practical hurdles – few individuals today are ready to hand an AI agent the keys to sensitive systems or credit cards without tight oversight. Those applications will come, but perhaps more slowly than some headlines suggest, due to understandable security and trust concerns.
Agent to Agent (A2A) is another emerging standard. This focuses more on allowing agents to discover each other and delegate tasks to each other. The key difference for our knowledge sharing example is that MCP shares ‘the knowledge’, whereas A2A takes on the task and returns the result. This is a subtle difference, but huge in terms of impact. A2A would offer experts the ability to limit access to their data but also take on the LLM costs to respond. They are also providing the answer, therefore potentially the responsibility for it being correct. Think of MCP being a library that lends a book, you can use that knowledge as you please in the report you are writing and so you are responsible for what you write. A2A is more like asking a book a question for your report and you accepting and using the answer verbatim.
MCP and A2A makes it possible for anyone to create and share capabilities that adds new expertise or skills to an AI Agent. We believe this knowledge-centric use of MCP and A2A will be a game-changer for expert advice in the B2B world. Instead of relying only on a general LLM, you can give your AI a “brain transplant” with domain-specific knowledge by plugging in the right modules. This opens the door for consultancies and expert firms to package their know-how as AI accessible services.
An InfoSec Advisor Example
To make this concrete, imagine you need advice on information security (InfoSec) policies for your company.
Today, you have a few options:
- Hire a person or consultancy: Bring in an InfoSec expert (as an employee or contractor) to assess your needs. The quality of advice can vary – some consultants are excellent, others less so, and you might not know which you’ll get. It’s also time-consuming and expensive.
- Use generic AI and hope for the best: You could ask a standard chatbot (ChatGPT, Claude, etc.) “What are best practices for enterprise data encryption?” and get an answer. But as discussed, it might be hit-or-miss. The guidance will be based on whatever the model absorbed during training (which could include outdated or low-quality info), and there’s a risk it omits critical nuances or even suggests something insecure. You would have to double-check everything.
Now envision a third option: open your AI assistant and add remote agent called “InfoSec Expert by ABC Security.” In this scenario, a reputable security firm (let’s call it ABC Security) has created a specialised A2A agent for answering InfoSec questions. This server is essentially an AI-ready knowledge base maintained by their top security consultants. ABC Security has distilled its expertise – up-to-date best practice guides, policy templates, incident case studies, and so on – into documents and data structures optimised for an LLM to use. They haven’t simply dumped a bunch of PDFs into a database; they’ve carefully curated and structured the content so the AI can easily draw on it to produce grounded answers. They also keep this knowledge base regularly updated with the latest developments.
If you connect the “InfoSec Expert” agent to your AI, suddenly your chatbot gains a trusted, focused repository of security knowledge to reference. When you ask a question, the AI’s answer will be backed by that vetted source, rather than by whatever happened to be in its general training data. The advice you get should be far more reliable and specific. It’s like having a top-tier security consultant whispering in the AI’s ear, making sure the response aligns with industry best practices as defined by ABC Security’s experts.
This approach directly tackles the earlier challenges. Hallucinations are minimised because the AI’s responses are grounded in a factual, company-provided knowledge base (a form of retrieval augmentation). If a detail isn’t in ABC Security’s content, the AI is less likely to invent it. Outdated information is less likely because the firm is actively maintaining its repository – it’s not frozen in 2021 like the AI’s base model might be but continuously refreshed. And quality is higher because the source material isn’t the wild internet, it’s vetted, authoritative content from a provider whose business and reputation depend on accuracy. In essence, you get the convenience and speed of an AI, combined with the credibility of a domain expert.
The Power of Curated Intellectual Property
The emergence of these expert AI modules shifts the value in the consulting world toward those who hold high-quality intellectual property. In the future, successful consultancies or specialist firms will be the ones that can translate their hard-won expertise into AI-consumable formats (and keep it updated). The strength of an AI advisory service like our hypothetical InfoSec agent comes entirely from the quality of its content. It’s not about having a more eloquent or “smarter” base model – the underlying LLM might be a commodity – it’s about the proprietary insights and guidance that the model is grounded in.
This gives companies with deep domain knowledge a new way to monetise it. Instead of (or in addition to) billing hourly for human advice, they can publish their expertise as an AI service. We can foresee a race to build the go-to AI advisor in each niche. We might see an “AI legal consultant” from a major law firm, an “AI supply chain guru” from a logistics expert, and so on. Companies will compete to have their AI-driven advisor be the most trusted in its category, much as they once competed for top rankings on Google search.
Interestingly, smaller, nimble players might have an edge here. At first, one might assume the big consulting powerhouses (McKinsey, BCG, etc.) would dominate this space thanks to their brand and vast knowledge libraries. But large firms could actually struggle to adapt quickly. The sheer breadth of their accumulated knowledge is actually a downside here – it’s like the internet: a huge dump of content with varying opinions and quality. Curation is critical. An AI advisor only needs the distilled best practices, not thousands of slides of mixed quality. A smaller firm that focuses on a specific domain can more easily curate a tight, high-quality knowledge base for that area, compared to a mega-firm with decades of diverse materials.
Thanks to user reviews and ratings, the best solutions – even from smaller players – will rise to the top. A boutique consultancy with an excellent AI advisor in a niche could outshine a larger firm’s offering if it consistently delivers superior value, because rapid feedback will quickly make clear which modules are truly helpful.
A New Business Market for AI-Powered Expertise
If this vision plays out, we’ll witness the rise of a marketplace for expert AI services. Instead of engaging a consulting firm for every need, a business leader might browse a catalogue of AI advisors: cybersecurity, HR policy, finance, supply chain optimisation, you name it. Organisations will race to publish high-quality agents to capture these opportunities, just as companies once raced to get their websites to rank on Google. In the early days of search engines, everyone focused on SEO – getting to the top of the results. In the near future, the focus might be “AI optimisation” – making sure your expert AI module is the one companies prefer and trust.
There are a few possibilities emerging for monetisation. One approach is for providers to charge per use of their agent via API keys, though managing many separate API subscriptions would be cumbersome in practice. More likely, we’ll see integrated marketplaces (or app stores) that bring many expert modules together under one roof. In fact, early signs of this are appearing. Platforms where companies can publish their MCP servers with descriptions and ratings, and users can subscribe or pay as they go to access them, with the marketplace handling billing and access control. A2A will no doubt follow suit as demand grows.
No matter the mechanism, providers will expect a return on their investment in creating these AI services. User feedback will be crucial. Just as app stores rely on downloads, ratings, and reviews to signal quality, an AI advisor marketplace will use similar indicators. Those metrics will help businesses figure out which AI advisors are worth paying for. Marketplaces are already taking steps to ensure quality and security – verifying providers and benchmarking server performance. Maintaining the knowledge base will also be critical: if an expert firm stops updating its content, its AI’s advice will quickly become stale and users will drop off. This creates a strong incentive for continuous improvement, which ultimately benefits users.
Will people pay for AI-delivered expertise? We believe yes, if it demonstrably saves time and money. Organisations already invest heavily in consultants; if an AI service can do 80% of a consultant’s work at a fraction of the cost, it will be attractive. For example, instead of paying a lawyer or advisor hundreds of dollars per hour to draft a compliance policy, a manager might use an AI compliance assistant for pennies per query and only involve human experts for final reviews. That kind of efficiency is hard to ignore.
Importantly, this isn’t about replacing human experts entirely. It’s about augmenting them and handling routine advisory tasks at scale. Human consultants and internal specialists will shift to higher-level work – defining strategy, handling complex or sensitive cases, providing the personal touch in client interactions – while AI does the heavy lifting of research and initial drafting. Consultants can even use these AI tools themselves to boost productivity (many already are). In fact, rather than diminishing the consulting profession, this technology might elevate the baseline of quality. Low-level, generic advice can be handled by machines, raising the bar for what human consultants focus on: more creative, nuanced, strategic problem-solving.
Conclusion
The future of expert advice in the age of AI is bright, provided we harness these technologies correctly. LLMs and apps like ChatGPT have opened the door, showing that instant, conversational guidance is possible – but also revealing the pitfalls of ungrounded answers. The rise of MCP and A2A (and similar approaches) signals a new phase: one where we can anchor AI in the rich, up-to-date knowledge of real experts. For consultancies and B2B service providers, this is both a challenge and an opportunity. Those who proactively turn their intellectual capital into AI-accessible modules will gain a new distribution channel and product. Those who cling solely to traditional, human-only consulting may find themselves outpaced by competitors who can deliver hybrid AI-human services faster and more cost-effectively.
There are many unanswered questions. Can MCP Servers secure IP? Serving content to another LLM for it to interpret does not mean the MCP Server is giving the answer, it is only enabling it. The clients LLM will still be responsible for the quality of the answer. What will people be willing to pay? Could this become another Spotify where artists get a fraction of a penny per stream?
Will a chatbot ever fully replace a top-tier consultant? For complex, high-stakes decisions, probably not – human experience, judgment, and empathy will remain invaluable. But for a wide range of routine questions and preliminary analysis, AI advisors will handle the load. And many clients will be happy to pay for that, especially when it comes packaged with the credibility of a known expert source. In effect, we’re looking at a democratisation of expertise: the best advice, distilled from the best experts, available on-demand to anyone who needs it – not just those who can afford large consulting contracts.
Consulting firms, large and small, should view this future with optimism. There’s a chance to reach more customers (including smaller clients that previously wouldn’t engage consultants) by offering scalable AI versions of their knowledge. The firms that thrive will be those that strike the right balance between technology and human touch, using AI where it adds value and humans where it matters most.
In the end, people will always seek expert advice. The form that advice takes is what will evolve. Tomorrow’s “consultant” might sometimes be a digital assistant backed by the collective wisdom of many professionals. Rather than eliminating the role of experts, such AI assistants will amplify their reach. The future of expert advice isn’t a zero-sum game of AI versus humans – it’s a symbiosis. By combining the depth of human expertise with the speed and scale of AI, we can solve problems faster and make high-quality guidance available when and where it’s needed. That is a future worth embracing.
