Managing AI Teammates On Engineering Teams

Sergei (00:00): Welcome to Clutch Conversations, the series where we dive into the critical challenges shaping businesses today. I'm Sergei, the SVP of development at Clutch and today we're talking about my favorite topic and something that I think you guys have heard of – it's called AI. But not just generally speaking, but we're breaking down how to manage AI in your engineering team. I'm joined today by Shanal Aggarwal, the chief commercial and customer success officer at Tech Ahead. They are a AI native app and enterprise software development company. So before we get started, Shanal, if you could tell me a little bit about Tech Ahead, maybe some of the exciting things that you guys have going on.

Shanal (00:47): Yeah, absolutely, Sergei. Very excited to be here. So we at Tech Ahead are at the forefront of building AI native applications and platforms and solving some of the most complex problems, which are really this with all these AI tools, we have an abundant supply of software and we have an abundant supply of ideas and we have abundant resources at our disposal to execute those ideas. But it is really hard to figure out what to execute, what ideas are good, what ideas are bad, where to put your energy with, to put your resources. And then it's also even difficult to ship these ideas into production at scale responsibly with the right guardrails and governance. So that's what we do at TikCare, helping enterprises navigate those complexities, those challenges right from ideation to discovery and strategizing the whole roadmap and transformation roadmap of how to best leverage AI because we hear companies approaching this, "Hey, we have an AI roadmap." Probably that's not the best way to look about it.

(02:04): It's the better way to think about it, what is our business roadmap and how AI can fit in and how we can leverage AI to deliver some really meaningful ROI and solutions internally as well as to their customers. So that's where we act as trusted partners for our clients. We are supporting them right through that strategy to execution and then shipping those products into production, ensuring they have those right guardrails in place.

How Tech Ahead Helps Companies Find Their AI Starting Point

Sergei (02:38): I think a lot of people probably would love to hear what you said at the beginning where it's like tech is getting cheaper to make. There's so many ideas. Where do I start? A lot of companies have struggles with, I guess, what are the business use cases? Is my infrastructure ready? If it is ready, who's involved? And I think it keeps a lot of people up at night, especially if there might be some misalignment internally. Where do you guys usually step into these roles as a partner for companies that are like, "Hey, I just don't know where to go."

Shanal (03:10): Yeah, that's a question that we hear a lot and that's where exactly companies approach us that, ‘hey, this is what we are trying to do and we don't know.’ And sometimes the ideas are even vague, the problem statements are not very clear. So we definitely try to jump in and understand the entire business landscape even before talking about a particular problem statement or understanding and jumping into solving a particular problem. So firstly, understanding what industry they are operating in, what are their constraints? Are they limited by data or do they have data at their disposal to build those systems and solutions or intelligence that they are trying to accomplish? And then identifying some of those key priorities, some of those key things which can deliver clear tangible results. Because what happens in some of these discussions is you come up with some really cool ideas, but at the end of the day, when you start thinking in terms of scaling it and deploying it across the organization, that's where you realize that these ideas are not scalable enough or are not actually solving the problems probably your internal teams are facing or the customers you face have.

(04:36): So understanding the business landscape, the target audience, who the users are going to be, what their pain points are, solving their primary goals for their primary goals and secondary goals is important. And then getting into the infrastructure side of things, which is data, what applications, tools they are currently using, how we can connect all these pieces together. And then later comes all these technology LLM models which are essentially enablers and how we can best configure them to deliver those outcomes they are looking for.

How Tech Ahead Uses AI Internally

Sergei (05:17): I love that word, enabler. So you obviously are an enabler for all these partners that you work with. I got to ask internally at Tech Ahead, what's maybe something that you guys have started using for yourselves? Is there an enablement that you guys have, whether it's engineering or sales process, where have you guys started? And I think some people would like to hear maybe how you even got to that problem and how you solved it.

Shanal (05:45): Yeah, absolutely. So I think as a technology team, we started around the 2009 era when smartphones were coming up and we always had the snack of picking up on emerging tech stack early and trying to understand how we can leverage that to improve our efficiency productivity and then offer the same to our partners, clients and customers. And with Generative AI too, as soon as ChatGPT launched, we started using these LLM models, we started not just with these frontier models by OpenAI and Tropic and all, but also started leveraging some open source models and building our own systems and tools around it. And some of the things which we did is to validate our ideas quickly, fail fast, learn from them, iterate, and then try and scale what is working. Because if you try to come up with a five-year roadmap, it's very difficult. The curve is so steep and so much is happening, you cannot run with it very fast.

(07:03): So right now, coming back to your question, I would say we are 70, 80% there in terms of doing entire engineering through AI tools, leveraging all these AI models to achieve what we call as AI native engineering.

(07:27): Also in our marketing and sales processes, or even in our project management and delivery operations, everything is something we don't really treat AI as a junior engineer per se because I think that's a very narrow lens of looking at it. If you configure it well, if you orchestrate all these AI agents well and you give them good guardrails, that's where the magic happens and that's what we try to do across all different units.

Is the Junior Developer Role at Risk? AI as a Multiplier

Sergei (07:58): It's funny, you said a five-year roadmap, right? It's hard to do a fear roadmap when the tech changes next week. Everything- Every

Shanal (08:07): Night. Every other day you wake up with something new notification or OpenAI did this, Anthropic did this.

Sergei (08:15): It's funny, you mentioned junior developers. You don't view them as a junior developer. What about the other side of it? Do you view the junior developer role as a risk in the future and then senior developers can manage these agents and systems better? There's arguments for both. I would love to hear maybe where you stand on that.

Shanal (08:34): Yeah, I think we have a slightly different take at it. We believe AI is bridging the gap between senior versus junior and AI is really a multiplier. So if you are, let's say in programming you are on a scale of one to 10, you are at five. And if you are using or probably at using AI, you are at five and if we assume AI is 10, so five into 10 is your outcome will be 50. And if somebody is at 10 out of 10 and they are using 10, so then 10 into 10 is 100. And we have seen some really smart people even early in their career. Cursor CEO, he's just 25 year old and he has the biggest AI startup in the world which was recently acquired by SpaceX. So I don't think experience is a barometer anymore and we don't treat that way at Tech Ahead.

(09:42): What we try to do is empower everyone to use these tools and share some best practices if some team achieve more productivity or efficiency by using AI in a certain way by writing through those markdown instructions or creating those skillsets, we definitely share that knowledge across, but we try to keep that flat structure, but within that it's more of giving ownership and accountability that who is that final approver and who is responsible for shipping it to the production, things like that. That's where I think still having somebody more experienced who has handled large scale deployments and understands what are some of the challenges, what happens when you are launching it for a million users and applications. So that's where I think experience matters, that's where system thinking matters. But for most of the work, I think it has really closed a gap between senior and junior.

Accountability, Human-in-the-Loop & CI/CD Guardrails

Sergei (10:52): I know we're talking about tech, but I love what you said about empowerment because it's not just about tech. It's anybody who's willing to learn this. I think it's a multiplier. And I think if somebody just goes at it, I love how you said there needs to be accountability. I want to ask a question about that in a second where there's somebody usually thinking on a systems approach. So if this junior developer does things 10 times faster, there's still somebody who has to check, but that developer now has the opportunity to 10X themselves and I think they become more valuable in an organization. I think a lot of people, there's headlines that are afraid of, well, all junior developers will be gone. I think they'll just become more productive. I don't know if you agree with that, but if they do, and I would love to hear from your side on Tech Ahead, the accountability layer, who is the person that checks the system?

(11:43): Because AI can create a lot of good stuff, but it can also create a lot of problems and bugs. So somebody's got, and how do you guys manage that?

Shanal (11:50): So yeah, that's a great question, and that's in fact one of the biggest problems what typically we see when even companies approach us, they have these pilots and POCs and systems that they have already developed, but they don't know what they don't know. And even AI doesn't know if you don't give it the context or if you don't ask the right question or you don't configure it for solving a particular issue, it won't solve that automatically. So we haven't, I guess, achieved that level of intelligence yet. So that's where having a human in the loop is extremely important and creating those structures where you have different environments built in for your standard CI/CD operations from dev staging production and having those right pipelines in place and having some sort of manual reviews along with automated reviews as well. So one of the cool things with AI is you can configure their role.

(13:02): So you give them the right guardrails and instructions and they can even play that role of auditor because even as humans, you might miss certain things, but if you share the right context and you structure it well, then it can perform those tasks as well on your behalf. And it's just setting it up the right way and then giving it the right guidelines. So we have this hybrid approach when it comes to the auditing side of things and approval side of things. We definitely don't want to have people reviewing thousands and lines of code every day because again, that is also becoming a big challenge. You can't really ask a human to review millions of lines of core, which AI is spinning out. So you need to have an LLM as a judge, you have to have some AI agents to solve that problem.

(14:09): And essentially it's a machine tackling the other machine at that speed.

Why Guardrails and Markdown Instructions Are the Foundation

Sergei (14:14): Some people might hear it and say, "Oh, CI/CD pipeline and agents reviewing code." And it just seems like it's overwhelming. But I think what you said before that was about structure and guardrails. Do you think that's where people should start? Because honestly, that might be the foundation of everything else will just work on top of it if you do it correctly. Do you believe that?

Shanal (14:33): Yeah, absolutely. And in fact, internally too, as well as what we have seen across the industry, high-performing engineering teams or AI-native engineering teams, they spend less time writing codes, but they spend more time writing all these markdown instructions or guardrails and deciding which agent has what level of autonomy do they have access to the production server or do they have just access to your staging environment where all those AI agents can make those changes? So I think having that clarity is really key to make sure what you are doing and then having some sort of observability and traceability in this entire process is also key and important because otherwise, even after writing those instructions and structures, AI has this habit of losing context and hallucinating, although there are systems and these models are getting better at it, but still there are these inherent issues within the technology for which you have to have those strong observable dashboards in place.

Sergei (15:53): For a lot of people who are going to listen to this too, it's markdown files are more powerful than people imagine. I view markdown files. I have three kids and if I just tell them to go draw with chalk, my whole house might be painted. But if I say go draw chalk on the driveway below the ... That's what a markdown file is. It's just context for the agent so that it doesn't hallucinate or screw things up. And I think people would need to start and look there. So I like what you said there.

Shanal (16:21): Yeah, absolutely.

Treating AI as a Team Member: Automation With a Human Hybrid

Sergei (16:24): So do you vie AI as a team member and then what kind of automation with that team member do you have? Do you have automated things, workflows, or do you do primarily what you said where it does it, but then a human has to check?

Shanal (16:39): No, so we follow a hybrid approach where there are certain human checkpoints, especially when we are dealing with regulated industries or we are dealing with production level databases and deployments, but then various things can be automated. And again, that's where you have to define that structure upfront, what level of autonomy you want to give to a particular AI agent or a tool you are using and that is where you have to, before you jump into designing and developing these systems, having those structures and guardrails in place and configurations, that's important and deciding which processes you want to automate. For example, if there is a customer support request, if there is a pug that is there, so the system can flag it, it can automatically create a Jira ticket. You don't really require a human going in and copying that from one tool to another and assigning it to.

(17:51): If you have the right structure, it can create that Jira ticket, it can assign those tickets to developers who are working on it. It can even close the status of that ticket once we have, okay, this code was fixed, the PR was approved or must, and then you can also build your own tooling or you can leverage some platforms which are already out there to create automated test scripts that can check your entire code and you don't really have to have a human in the loop for that process.

Where to Start: Automating Simple, Repetitive Tasks First

Sergei (18:30): I think what you're getting at, and correct me if I'm wrong, it's people that come to you or want to automate, they shouldn't be thinking about just automating everything. They need to start with certain business processes. Jira was a great one you mentioned because we even have a chatbot called DevBuddy where you can just create, close the task, add a pull request, everything from Slack. We don't even have to go into anything anymore. I'm surprised it's not tired how many times people tag it. But anyways, is that where you would tell people to start? Find things that are simple to automate or mundane, or do you try to attack both sides for some of your clients?

Shanal (19:08): For companies who are really starting their journey, I think if look at some of the mondayed repetitive tasks like sending announcements to your internal teams or just connecting your email with your Workspace agent or ChatGPT and getting a summary out of it using all these AI note takers, I think these are really ... Before AI, a lot of my time used to go in jotting down those notes and preparing those MOMs and even then you sometimes miss, but now I think that has become really easy. So I think starting with using some of these tools and exploring what their capabilities are and then trying to figure out what are some of the repetitive tasks and mundane tasks you have in your workflows that these tools can solve, that's a good starting point.

Claude vs. ChatGPT & Final Thoughts

Sergei (20:07): It's going to be a funny solve in five years because now it's so easy to create with OnePrompt, a 33-page document, every company now has six million documents that are just floating around everywhere. So I got to ask, and before we wrap up today, could be controversial, Claude or ChatGPT, if you had to pick today?

Shanal (20:29): The 5.5 model is really great in comparison to Opus 4.7, Mythos is not out in public domain yet, but on certain tasks and parameters, ChatGPT, ChatGPT 5.5 model does really good work, but I would still prefer Claude for coding tasks. But from a cost standpoint too, I think 5.5 is more token efficient. It burns less tokens, whereas Cloud, every hour or so you are out of limit.

Sergei (21:16): Yeah. And it's funny because even Gemini and Grock are moving up. And so I think if companies don't have the internal resources, finding a partner like you who can ... Sometimes you use all of them for different tasks inside of the company and that's a totally normal way of working. Well, I want to thank you for your time. Where can people find you? Is there a social website?

Shanal (21:40): Yeah, you can find us at techahead.com and you can find us with just take ahead on LinkedIn or Instagram or YouTube. So we have very active channels over there. And if you are interested in exploring how AI can be leveraged to transform your business operations or improve your productivity or HerAn's customer support and things of that nature, you can reach out to us and we can help you with that.

Sergei (22:15): Yeah. Thank you, Shanal, for your time today. I really enjoyed the conversation.

Shanal (22:19): Yeah. Thank you,
 

About the Author

Sergei Dubograev VP of Development at Clutch

Sergei Dubograev is the VP of Development at Clutch, the leading global marketplace of B2B service providers. With a proven track record of driving innovation and delivering exceptional results, Sergei has led cross-functional teams to create and launch products successfully in-market. His strategic mindset, coupled with his ability to identify industry trends and customer needs, has consistently propelled organizations to new heights of success. At Clutch, he oversees the strategic planning, execution, and optimization of the company's development portfolio, ensuring alignment with business objectives. 

See full profile