Every few days, another AI agent promises to change how we work. For a marketing leader, the temptation is obvious: connect the database, Shopify, inventory systems and cloud infrastructure, add a few agents, and start transforming the business.
But should you build that capability yourself? And if you choose a platform, what should you still own?
These were the questions I explored with Siddharth Gopalakrishnan on Dilse Omni Talks. What emerged was a practical masterclass in agentic AI adoption, grounded in team capability, proprietary customer knowledge and realistic expectations.
About the Guest: Siddharth Gopalakrishnan is the COO of Netcore, a leading customer engagement and marketing technology platform, where he drives strategy on AI-led marketing transformation and enterprise partnerships, including Netcore's collaboration with Google Cloud.
My biggest learning was simple: build versus buy is not a permanent choice between two camps. It is a journey, and different parts of your business may need different answers.
Table of Contents
1. Start With Your Team, Not the Latest Agent
2. Own the Intelligence That Makes Your Business Different
3. A Partner Can Be the Starting Point, Not the Final Destination
4. Marketing Leaders Need Technical Fluency
5. Do Not Oversell Agentic AI Inside Your Company
6. Content Remains a Difficult Frontier
7. Agentic Marketing Is Between Hype and Real Impact
8. The Advantage Goes to Businesses That Keep Learning
1. Start With Your Team, Not the Latest Agent
When I asked who should take the DIY route, Siddharth began with organisational readiness rather than technology.
His example was a business with a 20 to 50 member team that is already analytics-savvy, understands data architecture and regularly builds new capabilities. For such a company, building its own AI agents can make sense.
The important qualifier is not simply the headcount. It is what that team can already do. A company with an established building culture starts from a very different position from one still developing its data capabilities.
Even then, Siddharth was not advocating building everything. His advice was to develop the agents that matter to your business and use platforms for specialist domain agents that do not need to be reinvented.
That distinction matters because a company needs more than marketing automation. The architecture must serve the business, not merely add another marketing tool.
A Build, Buy, Collaborate Decision Framework
I would organise Siddharth's guidance into three decision cards. This is my synthesis of our discussion, rather than a named framework he presented.
BUILD: You have capability and distinctive business knowledge. Your team understands analytics and data architecture, and the agent relies on intelligence that differentiates your company.
BUY OR PARTNER: You need specialist capability or a faster starting point. Use a platform or technology partner where rebuilding adds little value or your internal team is not ready.
COLLABORATE: You need the whole workflow to work. Connect internally built agents with platform agents so that together they solve the business problem.
Team readiness → Selective ownership → Specialist support → Agent collaboration
The question is not, “Can we build an agent?” It is, “Which capability should we build, and how will it work with everything else?”
2. Own the Intelligence That Makes Your Business Different
Our discussion became particularly concrete around next-best-product recommendations.
A recommendation engine should not look only at the product catalogue. It may also need to understand customer behaviour, previous purchases and the customer's life stage. A business can hold a much richer understanding of its customers than it chooses to make available to an external marketing platform.
That makes the recommendation engine a strong candidate for proprietary development when the capability exists internally.
But identifying the next-best offer is only half the challenge. Delivering it at the right moment, when it is relevant to the customer, requires another set of capabilities. This is where Siddharth positioned Netcore's marketing technology agents.
I found this separation useful:
Decision intelligence: What should this customer be offered, based on what the business knows?
Engagement intelligence: How do we reach that customer at the right moment?
Collaboration: How do these capabilities work together to produce a useful customer interaction?
This is a more productive way to think about agentic AI than counting how many agents a company has deployed. A collection of agents is not automatically a functioning business solution.
3. A Partner Can Be the Starting Point, Not the Final Destination
What if the analytics team, architecture expertise and development capacity do not exist yet?
Siddharth's recommendation was to begin with a partner. Get something working, demonstrate early success and use that evidence to make the case for further investment.
That success can help secure budget in the next annual operating plan and support the development of an internal team.
“Teams built in-house are the best teams for things like this.”
Siddharth Gopalakrishnan
The nuance behind that statement is important. His preference for internal capability did not mean every company should start there. It meant that the starting point and the longer-term destination can be different.
As I summarised during our conversation, a business can start with a partner and bring more capability in-house as scale and investment justify it.
Neither route is effortless. AI skills are expensive, and retaining the people who have them is difficult. The build-versus-buy decision must account for the team you can sustain, not just the team you would like to hire.
Partner → Demonstrate success → Secure investment → Build internal capability
4. Marketing Leaders Need Technical Fluency
I then asked a slightly uncomfortable question: do marketers understand enough technology to lead this shift?
Siddharth responded with humour, but also genuine optimism. He has been impressed by marketers, including those with more traditional backgrounds, who have gone deep into technology and learned through hands-on experience.
His conclusion was that a tech-first marketing team now has considerable value.
I see two learning journeys converging. Brand marketers with 10 or 15 years of experience are learning and unlearning as technology changes their work. Growth marketers are discovering the importance of content, storytelling and virality.
AI sits at that intersection. A brand marketer may bring stronger judgement to content creation. A growth marketer may bring deeper expertise in conversion rate optimisation. Both need to understand how technology can support their craft.
The opportunity is not to replace marketing judgement with technical enthusiasm. It is to combine them.
5. Do Not Oversell Agentic AI Inside Your Company
Marketers are good at selling ideas. With agentic AI, that strength can become a problem.
I raised the risk of promising too much internally. A compelling demonstration creates excitement. We imagine that a tool will change everything. Then we use it on real work and discover its limitations.
Eventually, the assessment becomes more grounded: perhaps it does not transform everything, but it does improve something we previously did manually.
That is a perfectly useful outcome. It only feels disappointing when the original promise was unrealistic.
My takeaway is to establish three expectations before an initiative begins:
A specific problem: Be clear about what the agent is intended to improve.
Room for failure: Treat experimentation as learning, not a guaranteed success story.
A credible comparison: Judge the result against the existing way of working, not the most exciting demonstration.
This is how we protect confidence in the programme while still giving teams permission to experiment.
6. Content Remains a Difficult Frontier
Siddharth was especially cautious about AI-generated content.
Some premium brands want close control over every element of communication that leaves the organisation. That is understandable. Their standards affect how freely they can use AI content tools, even when they are adopting AI elsewhere.
His assessment was that content remained one of the toughest areas to get right. He was also reluctant to predict how quickly that would change.
Meanwhile, the demand for content has grown dramatically.
I recalled the earlier model: one major summer advertisement featuring Shah Rukh Khan or Salman Khan, followed by multiple edits that could carry a season. Now a brand may need to ship 50 content pieces a month across Instagram, Meta, YouTube and other destinations, with personalisation adding further complexity.
But content volume is only one part of the problem. The recurring decision is:
What content → For whom → On which channel → At what moment
Generating more material does not, by itself, answer those questions. Creative judgement and engagement decisions still have to come together.
7. Agentic Marketing Is Between Hype and Real Impact
When I brought up the Gartner hype cycle, Siddharth placed agentic marketing somewhere between inflated expectations and genuine arrival.
The hype is the belief that it will solve everything. The progress is that some models are working well, and the wider conversation has made organisational change easier to initiate.
Demonstrating results, however, still takes time.
“Experiments are at its peak. Real impact stories are still few.”
Siddharth Gopalakrishnan
At the time of our conversation, he saw the following six to twelve months as an important period for proving impact. Netcore's planned outreach, he explained, would increasingly centre on customers with demonstrated results.
The commercial tension was equally revealing: excitement was high, model costs were substantial, but there was little evidence of customers being willing to pay a premium simply for AI interventions.
My response was that payment would follow outcomes. Siddharth agreed. More credible impact stories could support the next wave of adoption, while lower costs could make the economics more sustainable.
8. The Advantage Goes to Businesses That Keep Learning
Towards the end, I used an autonomous-driving analogy: the advantage goes to the system accumulating more kilometres of experience.
For brands, the equivalent is running experiments, learning from failure and trying again. Beginner's luck is not a dependable AI strategy.
Siddharth closed with a useful warning: getting agents to work well can take far more effort than getting an AI model to work. His description of “10x” the effort was a reminder not to confuse an impressive model response with a dependable agentic workflow.
I left with a clearer principle: build selectively, partner pragmatically, connect the capabilities and earn the right to scale through results.
I am Saurabh Agrawal and we come with a new episode on Dilse omni talks every fortnight and cover different aspect of omnichannel with amazing speakers.

















