AGENCY GROWTH CLUB
Travis Tallent (DayNova AI ) on Rolling Out AI Without Losing Your Team
Guest: Travis Tallent
Guest: Travis Tallent
Travis Tallent, Founder of DayNova AI and former Head of AI at Brainlabs, learned the hard way that telling a large team to “use AI” is not an adoption strategy.
The rollout needs a vision that explains what remains distinctly human, what work can be supported by systems and agents, which team members have capacity to learn, and how people will be protected from unnecessary confusion or fear.
In this Agency Growth Club conversation, Travis shares the practical systems he uses for marketing and leadership, DayNova’s strategy-build-embed model, his search and AI-cost outlook, and why communication and influence still matter when AI becomes baseline capability.
The following chapters are editorial timestamp labels based on the supplied transcript. The public YouTube metadata returned the episode title but did not expose a chapter list, so these labels should be reviewed against the final uploaded video before publication.
Travis says the directive behind an earlier rollout across a roughly 90-person team was simple: get everyone to use AI. The problem was that the instruction came without enough training, framing, or room for questions. In his account, the implicit “use AI or perish” message made people anxious about their roles and unsure how the fast-changing tools applied to their day-to-day work.
His alternative is an AI vision. A leader should say openly that the organisation does not know everything about AI yet, identify valuable use cases together, and be explicit about where people should invest their time. Travis gives strategy, human relationships, and polished client presentations as examples of work he wants people to deepen. He contrasts those with repetitive work—such as internal linking, content briefs, and outbound-PR support—that agents and systems may be able to assist.
“Create, instead of an AI mandate, an AI vision.” — Travis Tallent
The phrase is not a call to avoid standards. It is a call to replace an ultimatum with direction: what problem are we trying to solve, what quality safeguards matter, what is changing, what is not changing, and how can people contribute to the design of the new workflow?
Travis says some people will naturally become early AI champions while others will be more cautious. He does not see that difference as a problem. The management issue appears when champions receive no boundaries and several people build nearly identical systems in parallel. That wastes time, creates inconsistent methods, and makes it harder to identify what actually works.
His response is explicit ownership. One person or group might build an internal-linking agent; another might focus on a content-brief system; another may explore an outbound-PR workflow. The goal is a coordinated portfolio of experiments rather than a collection of isolated tools that nobody maintains.
Actionable advice: Build an adoption backlog before launching a tool race. Define the use case, intended user, data boundary, quality bar, owner, review date, and decision about whether the work will be scaled, changed, or stopped. That structure preserves room for experimentation while reducing duplication.
For marketers, Travis starts with repeated processes. Internal linking, metadata, and content briefs often contain a sequence of steps that can be described, refined, and reused. He recommends thinking about the system from start to finish, including client or brand variations, then incorporating that system into the AI environment a team already uses. He mentions Claude skills and Markdown files used with ChatGPT as examples of portable formats.
The asset, in this view, is not one clever prompt. It is the documented process, unique standards, client constraints, examples, review checklist, and iteration logic that makes the work repeatable without making it generic.
For agency owners, Travis’s example is a scheduled “inbox sweep.” His system runs three times a day—at 8 a.m., 2 p.m., and 5 p.m.—to review inbox and Slack activity, surface what is new, and triage what needs a response. He says this reduces the cognitive burden of constantly scanning messages while switching among meetings and priorities.
Actionable advice: Start with a narrow recurring problem. Estimate how much time or context switching it creates, document the current steps, decide what data can safely be included, create a human review point, and test whether the system reduces genuine friction before extending it to other work.
Travis likes Obsidian for personal use, but says it is less suitable as a company-wide system when permissions are important. He notes that teams can build more elaborate technical workarounds, but that a system such as GitHub with a strong folder structure may be more practical for DayNova’s context. He also names Notion and Google Drive as potential alternatives.
The wider point is not that one software choice is universally best. It is that AI adoption magnifies a question businesses already need to answer: who can see what information, who may change a process, what context is appropriate for a tool, and where is the authoritative version of a system held?
This is why a “second brain” is not automatically an enterprise knowledge base. Individual convenience, version control, permissioning, compliance, retrieval, and governance need to be designed for the actual organisation and data involved.
Josh raises the appeal of leaderboards, incentives, and friendly competition. Travis agrees that gamification can be powerful, but adds a critical condition: it must match the team’s appetite and available time. In some circumstances, he suggests an invite-only challenge or an application process rather than a mandatory game for everyone.
The concern is operational rather than philosophical. A person already overwhelmed by client delivery may not benefit from an additional learning task. If the system is difficult to learn and not yet improving their output, a game can push work into evenings and make the adoption experience worse. Travis says AI has a real learning curve; treating it as effortless can alienate exactly the people who need support.
“I’m a big fan of the games, but with caveats.” — Travis Tallent
Travis recommends an initial assessment: do people feel threatened or excited by AI, and do they have capacity to explore it? The outcome should be a more deliberate rollout—one that gives interested people a structured way to experiment while giving overloaded people the time, guidance, and workload space they need.
Travis describes generative AI as probabilistic and prone to a “yes bias.” Understanding that tendency, he says, creates a healthier skepticism toward outputs. A system may produce fluent, agreeable material without reliably applying the organisation’s taste, standards, evidence, or unique method.
That leads to an architectural recommendation. Data retrieval, data warehousing, rules-based calculations, and other deterministic operations should remain deterministic. AI can then sit at an insight, synthesis, or ideation layer on top of structured inputs, where probabilistic reasoning is appropriate and human review can challenge it.
The final element is differentiation. Travis advises teams to identify what makes their process special, then make that explicit in their system. Otherwise, the model will supply generic defaults for outreach, internal linking, or other work—and the resulting output will resemble everyone else’s.
Actionable advice: Define the boundary before building. What must be exact? What can be suggestive? What requires an approved source? What needs a human reviewer? What specific principles or examples must the system follow? Those questions make a better design brief than “automate this with AI.”
Travis says DayNova begins by mapping AI strategy to a client’s existing operating model. This is meant to identify champions, constraints, bottlenecks, pressure points, and where a generic AI blueprint would fail to fit the organisation.
The second stage is building: selecting and developing automations that alleviate the relevant bottlenecks. The third is embedding: team coaching, training, change-management support, and—in the right situation—gamification. Travis says that the founding team’s neuroscience, coaching, product, data, and technical experience informs this more holistic approach.
He believes clients need all three elements to see a meaningful return from AI work. That is DayNova’s delivery philosophy, not an externally validated implementation formula. Its real value for readers is the sequence: understand work before automating it, build for a particular operating constraint, then invest in adoption rather than treating deployment as the end of the project.
Travis connects AI resistance to VUCA: volatility, uncertainty, complexity, and ambiguity. He says a major change can feel threatening to the human brain, and AI comes with unusually high stakes in people’s minds because tools evolve quickly and media coverage often emphasises risk to jobs and industries.
In his view, a leader may intend to say, “Use AI so we can stay competitive,” while a team member hears, “Use this technology or your job is at risk.” That interpretation increases cognitive load and can lead to worse work, not faster adoption. Travis says he has spoken with people in several industries who are exploring new employment because they feel leadership has not listened to their experience of AI change. This is his observation from conversations, not labour-market research.
The leadership task is to reduce VUCA where possible: state the business objective, name what is uncertain, clarify decision rights, explain the current effect on roles, create channels for feedback, and give employees room to learn. An AI vision becomes credible when it is accompanied by time, boundaries, and listening—not only a slogan.
Asked what he would say to people unhappy with slow or poor AI adoption at their employer, Travis warns against assuming that the grass is greener elsewhere. His phrase is that “the grass is green where you water it.” He encourages people to try to influence what they can in their current role before concluding that a new company will solve the issue.
If someone has genuinely reached the limit of their influence, he says it may be reasonable to explore other opportunities. Travis also recommends being clear about the real reason for leaving. If compensation is the main issue, a direct conversation with a manager may be worthwhile before an external search, because retaining a capable employee can be preferable to recruiting and training a replacement.
This is Travis’s career perspective, not a one-size-fits-all prescription. Individual circumstances, power dynamics, financial needs, and workplace culture vary. The transferable principle is to distinguish a solvable operating concern from a broader mismatch of role, values, opportunity, or trust.
Travis calls AI “table stakes.” His larger career differentiator is communication and influence: the ability to articulate why a strategy matters, persuade stakeholders to test an idea, explain why a campaign is or is not working, and help a group make a decision.
This matters because marketing work is not finished when an analysis is produced or a tool is used. Someone still needs to frame evidence, surface trade-offs, answer questions, gain alignment, and maintain a client or internal relationship through uncertainty. Travis believes AI can support this work, but does not remove the human need to be credible, clear, and influential.
Actionable advice: Practise explaining work at three levels: a one-sentence outcome statement, a concise decision brief, and a detailed evidence-based account for specialists. AI can help rehearse and refine the communication, but the underlying judgment, context, and accountability remain human responsibilities.
Travis expects search budgets to rise, in his view, because paid visibility may require more spend to achieve comparable outcomes while organic visibility now depends on more than a single website. He names Reddit, TikTok, Amazon, YouTube, and other surfaces alongside traditional search.
He also sees a budget-ownership problem. For example, YouTube investment may sit with a creative or video team even when the channel is increasingly relevant to discovery and search. His answer is search-first marketing: organise around how buyers search and choose, rather than treating channel silos as separate from visibility.
These are strategic forecasts from the episode, not budget guidance. The specific allocation that makes sense will depend on a brand’s audience, product, market, data, existing capability, commercial model, and measurement system.
Travis explicitly says he is not predicting the disappearance of all business categories. He thinks affiliate sites are in a difficult position, but says brands that sell products or services can still perform well. His distinction is between website traffic and business value: as Google and ChatGPT keep more comparison and transaction activity inside their own products, click-through traffic may become less reliable as the sole measure of success.
In his view, a brand can see lower website traffic while transactions or revenue remain stable or improve if it is effectively surfaced and chosen inside those platforms. He expects the old journey—search, click, compare on a website, convert—to become more complicated as comparison happens increasingly inside AI interfaces and search products.
That is a forward-looking operating view, not a claim that every search journey will work this way or a promise that lost traffic will be replaced by platform transactions. The practical message is to improve measurement beyond sessions and continue to understand how visibility, selection, conversion, revenue, and customer relationship connect across channels.
At the time of recording, Travis says DayNova was around a month and a half into its independent journey. He did not believe the right first move was a fully expanded “search everywhere” strategy. A new company has limited time, money, and capacity, and filling the calendar with discovery calls before delivery is ready can simply create a different bottleneck.
His approach is stair-stepped: begin with lower- and mid-funnel activity appropriate to the business’s maturity, then expand as the company has more capacity. Travis says one early client came from a relationship. He is also using LinkedIn automation to share helpful webinars and diagnostic tools instead of sending a pitch-first message.
He describes adjacent partnerships as another important route. The idea is to work with businesses that serve a similar client base but do not compete with DayNova’s strategic and automation work. He gives his educational work with Chatwalrus as an example, describing it as complementary: he teaches practical Claude and Gemini use, while the partner serves its own training role.
Actionable advice: A new consultancy does not need every channel active on day one. Choose the growth method that matches capacity, service readiness, and existing relationship equity. The test is not whether a strategy looks comprehensive; it is whether it creates the right conversations without compromising delivery.
Travis’s personal prediction is that routine AI cost will fall. He points to competition, lower-cost models, local-first deployments, and the ability to handle some tasks through conventional automation rather than a cloud model billed for every token. He believes businesses will increasingly reserve higher-cost frontier capability for work where superior research, coding, or high-intensity reasoning produces clearly observable value.
He also cautions against importing token-cost stories from developer-heavy technology companies into every organisation. In his view, a non-technical business may use substantial automation without approaching the spend profile of a company that runs intensive model workloads at scale.
This is not an AI-budget forecast or vendor-pricing recommendation. Cost depends on the model, provider, contract, task, usage volume, engineering decisions, data, local hardware, governance, and the value of the output. The useful planning question is not “will AI be free?” but “which work needs expensive capability, which work can use a cheaper system, and which work should remain conventional automation?”
Travis describes himself as privacy-first. His immediate recommendation is to review a tool’s settings and disable model training where the product and account type offer that control. That is a sensible prompt to investigate data handling, but it is not a universal guarantee. Organisations should verify current product documentation, enterprise terms, data-retention settings, access controls, and applicable security and compliance requirements before placing sensitive material in any system.
He also expects more interest in local-first models and conventional APIs where they fit the task. His concern is not merely external exposure. He says that feeding large amounts of unguarded context into a model can increase hallucinations, which may create operational problems even if the data stays within approved systems.
Travis’s priority is infrastructure: understand what creates risk for the business, then build around it. That includes clear decision rights, permission boundaries, appropriate data pipelines, and knowledge of who can view or change particular systems. Many organisations still need these foundations whether AI is involved or not.
“Be very skeptical of any computer output ever.” — Travis Tallent
Travis explicitly says he is not a company-valuation expert. His personal view is that some AI companies may fail and that some current valuations may be too high, but this does not mean the underlying AI technology or infrastructure disappears. He uses the dot-com era as an analogy: companies failed while internet adoption and the technology’s eventual importance continued to grow.
This is not investment guidance, a valuation assessment, or a prediction about any particular company. The editorial value is in the distinction itself: market cycles can affect individual businesses while broader technology adoption continues. Readers should avoid treating that observation as a basis for financial decisions.
An AI mandate is a blanket directive — “use AI or perish” — with no training or context, which tends to create fear rather than adoption. An AI vision is specific: it names which tasks should stay human-led and which should be handed to AI agents, giving a team direction instead of an ultimatum.
It can, but only for people who already have capacity to learn something new. For team members already stretched on delivery, gamification adds pressure rather than motivation — Travis recommends assessing individual capacity and sentiment before rolling out incentives.
No — Travis expects search budgets to rise, driven by expanded paid search placements and the need for organic visibility across more surfaces than a brand’s own website, including Reddit, TikTok, Amazon, and YouTube.
Travis expects the opposite — prices trending down, driven by sunk R&D costs, cheaper competing models, and a shift toward local-first, deterministic automation for simple tasks. Frontier models will stay expensive but reserved for high-leverage work.
Turning off model training in an AI tool’s settings — a one-time toggle that stops personal or company inputs from being used to train the underlying model.
VUCA stands for volatility, uncertainty, complexity, and ambiguity — a 1980s change-management framework describing how the brain reacts to disruption. Any large change, including AI adoption, triggers it; leadership’s role is to reduce VUCA through clarity, not add to it through vague mandates.
Communication and influence — specifically, the ability to explain and defend results to a client or stakeholder in a way that builds trust, not just deliver the results themselves.
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.
