AGENCY GROWTH CLUB
Kai Ali (Montaj Digital) on Why AI Will Cost More Than Your Team
Guest: Kaiyan Ali
Guest: Kaiyan Ali
Kai Ali, founder of Montaj Digital, argues that today’s AI subscriptions are subsidized by venture capital and will get more expensive as models improve, meaning the “cheap” AI replacing lower-salaried roles right now won’t stay cheap.
His fix isn’t to avoid AI, it’s to stop drinking from the same pond as everyone else: build one shared, contextual knowledge base your whole team can pull from, put human-in-the-loop checks on who can change it, and treat AI as a co-pilot rather than the pilot on anything with real risk attached.
Pulled from the transcript’s real per-line timestamps, matched to where each topic starts.
Josh introduces Kai as a close friend who has rebuilt his digital agency, Montaj Digital, around helping larger enterprise companies implement AI workflows. Kai describes the agency’s focus plainly: “We help service based businesses streamline their workflows using AI and automation to help them run more efficiently.”
Most businesses he works with either know AI matters but don’t know where to start, or have processes too weak for AI to add any real benefit, so Montaj Digital builds what Kai calls “private AI operating systems” designed to reclaim hours and protect margin.
Josh raises a pattern he’s noticed at AI events: the loudest people in the room often know the least, while the sharpest operators feel furthest behind. Kai confirms it from direct experience, at a London event with over 300 attendees across five AI implementation workshops, he polled the room at the start of each session on how well people felt they were keeping up. Almost everyone who said they felt on top of it came back afterward and admitted, “I thought I know about AI, but clearly I have no idea.”
Josh sets up the core problem of the episode: in a world where anyone can access the same powerful models, what actually differentiates a business? Kai’s answer is a “contextual layer”, the equivalent of strong IP, deep customer understanding, and command of your own business’s language, layered around a generic AI model so its output actually reflects your business rather than generic best practice.
The practical challenge, Kai explains, is that this context tends to end up scattered, some files in Google Drive, some in Notion, some in email threads, some in Slack. “To actually unlock a knowledge base, you really need to have a really sound operating system around the business.” It doesn’t matter which platform holds it, GitHub, Google Drive, One Drive, or Confluence (Kai mentions using it with a current enterprise client in Brussels), only that it has a clearly defined structure the whole team can pull from.
Actionable advice:
Kai flags a risk specific to larger organizations: if even 10 people are updating a shared knowledge base and leadership doesn’t agree with the direction that context is taking, that’s a dangerous position to be in. His fix is a regular review cycle, a daily or weekly summary of context changes sent to board level for approval, with the underlying database only updating once that approval is given.
Josh shares his own experience running a small team, his personal knowledge was stored entirely in Obsidian, and he assumed the team had absorbed the same understanding by osmosis. They hadn’t. After switching the team to Notion, Josh discovered on a strategy call that several team members didn’t even have basic tools like Gmail connected, despite him talking about AI constantly. “I’ve been talking about this for months. I was like, why have you not been able to do that?”
Kai says this is an extremely common story, including at a client doing hundreds of millions in annual revenue with a 35-person marketing team.
Everyone was told to have Claude Enterprise, but when the team split into departmental projects, some people had never used Claude beyond the basic chat interface and had never made a project folder. Kai’s suggested fix, first raised at a workshop the day before this recording: an “onboarding skill” that runs through every tool that should be connected, checks whether it is, and provides setup instructions if it isn’t, effectively onboarding a person’s AI the same way you’d onboard them as a team member.
Actionable advice:
Josh notes that ideas which used to take weeks or months to build now take a couple of evenings, and asks what Kai does with the time AI has bought back. Kai admits he still builds constantly and sometimes has to check himself on whether a project is actually important. His stance is nuanced: he recommends “sandbox time” to clients, giving a team half a day every week, fortnight, or month with no fixed objective other than solving a problem using AI, as one of the best ways to build real skills.
The risk, Kai says, is dopamine-driven building with no follow-through: “I do have a lot of friends who either are AI native or are learning a lot about AI that are just so addicted to the dopamine of building things that they’re not doing anything about it.” He references a contact’s line that distribution has always been king, and is now even more evident since anyone can build anything. For Kai personally, tangents that don’t go anywhere still sometimes spark an idea he acts on a month later.
Actionable advice:
Josh asks Kai to revisit a framework from his London keynote: taste curation, or how to become “AI-proof.” Kai frames the underlying problem as knowledge democratization, everyone can now access the same information, which creates two failure modes: people getting “fake dopamine” from building things that don’t matter, and people mistaking AI-sourced knowledge for genuine expertise. “If you don’t know anything and you learn everything from AI, you just assume that’s the truth”, even when a figure the AI gives you, like a salary range, might be off by a wide, indefensible margin.
His fix is building a “taste library”, deliberately seeking out where popular frameworks and books originally came from, rather than just consuming the popular summary everyone else is using. He put it as “don’t drink from the same pond”: everyone quoting the same handful of well-known creators (he names Alex Hormozi, Sabri Suby, and Daniel Priestley as examples he admires) ends up sharing the same recycled ideas, because they’re all working from the same source material. Going further back, to what those creators themselves learned from, is how you form an original point of view instead of adopting someone else’s without realizing it.
Kai ties this back to a trip to Cyprus, where every menu, billboard, and social post seemed to use the same recognizable AI-generated image style: “You can’t resonate with that, because business is and always will be human to human.”
Actionable advice:
Asked where he intentionally limits AI use to protect the human element, Kai draws a clear line: he’s happy to use AI for brainstorming, but when building a new workflow, marketing campaign, or offer that needs to carry real value, he tries not to start with AI at all. “I try to do as much of it with God’s gift in my head for as much of it as I can, because what that allows me to do is one, keep that human judgment, but two, formulate my own opinion so that I can use my own taste library, my own context, before I then go to AI to help me with that.” He also avoids AI entirely wherever being wrong carries a high risk of serious harm to the business.
He gives a concrete example: geometric design work, like PowerPoint decks, isn’t something he trusts to AI regardless of how much context it’s given, fine for internal reports, not for client-facing work. His team still uses AI image tools (he names Higgsfield, Nano Banana 2, and ChatGPT’s image generation as strong options) to brief and generate assets, but a human designer, one Josh’s recruitment work helped them find, and who has been with Montaj Digital for almost a year, remains in charge of the final creative direction, because his specific skill is invoking human emotion.
Kai frames the designer/AI relationship as a broader decision-making rule: AI is the co-pilot, the human is the pilot. In clear skies, low-stakes, routine situations, the co-pilot can run things largely on its own. The moment something goes seriously wrong, control needs to shift fully back to the human, the same way pilots train for hundreds of hours specifically because even a 0.1% error rate can risk lives.
Kai’s practical version: “If there’s no risk of any substantial impact, I’d be using all AI. If there’s some risk of AI getting me halfway there, but I definitely need a human to help mould that, I would be using the AI [with oversight]. If there’s so much risk of things going wrong… I’ll be using no AI whatsoever.”
Josh adds a related pet peeve, team members copy-pasting AI-written updates instead of communicating plainly, turning a short “this is done, this is why” message into an overwritten, AI-flavored paragraph.
Actionable advice:
Josh shares a live example: a PE-backed client with a Head of AI and 30 engineers at their disposal is looking to hire someone to “SaaSify” their agency model, someone who’s never had a shortage of ideas, only a shortage of resources. Kai says he’s seeing this trend everywhere, though it’s untested at scale: turning a repeatable service business into a software business isn’t new, it just used to require budgets only very large companies had.
He connects this directly back to hiring: “The companies looking to replace their people with AI are looking at it so wrong,” because current AI pricing is a loss-leader subsidized by venture capital, and API costs will only rise as models improve, eventually approaching or exceeding the cost of the lower-salaried roles people are trying to cut. Josh backs this up with a recruiter’s-eye view: he recounts pushing back on an audience member at Kai’s event who wanted to replace their “whole team” with AI, it turned out to be an offshore VA team doing genuinely repeatable tasks, which is a different question entirely from replacing skilled specialists.
Actionable advice:
Josh backs up Kai’s “wrong hire” argument with hard numbers from his own recruitment business: in the last 10 days, Search for Hire filled three search-marketing roles at £200k or more, two of them at £250k, all heavily weighted toward AI search skills. “If you’ve been doing SEO or you’re doing digital marketing and you don’t have a strong grasp on AI, you’re not getting past anybody’s filtering process.”
Kai extends the economics further: leaner enterprise teams are increasingly paying a premium for fewer, stronger people amplified by AI agents handling deterministic work, rather than employing large teams of mid-level hires.
He walks through the maths, a £250k hire supported by AI-driven deterministic task execution can outperform ten £50k hires costing £500k combined, even before accounting for reduced management overhead and organizational complexity. Whether the wider “SaaSify everything” trend holds up long-term, Kai says, depends on whether the SaaS companies buying at 20-40x valuations can actually generate returns in the next 5-10 years.
Actionable advice:
Kai recaps his Brussels keynote, delivered to a company generating hundreds of millions in annual revenue with 500 employees across sales, marketing, operations, and technical teams.
His core argument: now that everyone has access to the same AI models, the differentiator is an “intelligence layer” competitors can’t copy. The problem he identified is structural, senior leadership communicates well with each other, but departments only sync monthly, weekly, or even yearly, leaving each team’s context siloed from the others. A sales team’s insights about ideal customers, for example, often never reach the marketing team building the next campaign.
The best question from the audience came from a technically-minded new CEO who kept a personal knowledge base but didn’t want the wider team updating it freely, for fear board-level leadership would disagree with resulting changes.
Kai’s answer mirrors his advice for client-facing automated workflows: build in a human-in-the-loop layer. Redact sensitive information as standard practice, then route context changes through a named, siloed owner, marketing insights to the CMO, finance updates to the CFO, who approves or rejects updates before they reach the shared source of truth.
Kai also distinguishes between a personal knowledge base (one-to-ones, individual insights, personal frameworks) and a company knowledge base (tone of voice guidelines, avatar/ICP profiles, he describes one client’s “avatar battle cards” for four different products). Sales call transcripts, he suggests, should feed into the shared brain and be analyzed against those avatar profiles to surface real-time insight, for example, if five of the last seven sales calls raised the same budget objection, that’s content and messaging feedback the marketing team should see immediately, not next quarter.
Actionable advice:
Closing out, Josh asks what people ask Kai most often now that AI’s importance is no longer in question.
The answer, almost universally: “Where do I start?” Kai’s response is that AI and automation don’t work on top of poor process, the fix is stepping back to map out what the business actually does, task by task, before touching AI at all. “A job is just a series of tasks, and a task is just a series of actions.”
Once that’s mapped in a flowchart or written form, it becomes clear which tasks need AI, which need a human-plus-AI combination, and which need a human alone, and that mapping exercise is the literal first thing Montaj Digital does with every new client, before any AI conversation begins.
Actionable advice:
It’s the business-specific knowledge, tone of voice, and customer understanding layered around a generic AI model so its output reflects your actual business rather than generic best practice — the thing that lets two businesses using the same AI tools produce very different results.
Kai believes so — current AI pricing is subsidized by venture capital investment rather than reflecting true running costs (compute, energy, water), so API and subscription costs are expected to rise as models improve, potentially approaching or exceeding the cost of the lower-salaried roles they’re currently displacing.
Because if a role is repeatable and low-value enough that AI can fully take it over, Kai argues that was never the right hire for the business in the first place — the problem is the hiring decision, not the existence of AI.
It’s the practice of tracing popular frameworks, books, and creators back to their original sources rather than only consuming the popular secondhand summary — which helps you build an independent point of view instead of recycling the same recycled opinions everyone using AI is now producing.
He uses a pilot/co-pilot framework: low-risk, routine tasks can run mostly on AI; tasks where AI can get partway but need human judgment get AI-with-oversight; and anything with serious consequences if it goes wrong gets no AI at all.
Josh reports three £200k+ roles filled in 10 days (two at £250k), all requiring strong AI skills — suggesting that AI competency is becoming a baseline requirement that drives salaries up, rather than AI simply eliminating the roles.
Separate personal knowledge (individual notes, one-to-ones) from company knowledge (tone of voice, ICP, tested messaging), and route any proposed change to the company knowledge base through a named, responsible approver for that category — e.g. the CMO for marketing insights — before it updates the shared source of truth.
A structured checklist process that runs a team member (new or existing) through every tool and knowledge-base connection they should have, checks whether it’s set up, and provides instructions if it isn’t — effectively onboarding a person’s AI access the same way you’d onboard them as a new hire.
With 95% of roles filled within 25 days, our approach is built for pace and backed by a 1 year guarantee. Get access to the systems and frameworks that power elite SFH teams.
