Table of Contents
Every talent acquisition leader is hearing three pitches right now. Your ATS vendor says you already have everything you need. An engineer you know says, “I can build this with AI over the weekend.” And specialist vendors promise to automate everything and get you out of spreadsheets.
They can’t all be right. Or maybe they can, just not for every team.
That tension was the heart of GoodTime’s recent Spicy TA Q&A, where Hung Lee, founder of Recruiting Brainfood and a 25-year recruiting veteran, and Kyle Lagunas, industry analyst and founder of Kyle & Co, took on the build-versus-buy debate. GoodTime Senior Customer Success Manager Sejal Madhubhai moderated.
The hosts promised a fight. What emerged was more useful: a clear-eyed look at where AI-built tools create value, where they quietly create cost, and how to tell the difference. The short version? The question isn’t really build or buy. It’s scale, ownership, and whether your team can still work together.
Faster individuals don’t make a faster team
Start with the obvious: vibe coding works. A recruiter who builds a quick tool to kill a manual process gets real time back. Lee doesn’t dispute that. The problem shows up one level higher.
“Where it becomes challenging is converting this individual productivity into institutional value,” he said. “A bunch of people individually vibe coding their tools collectively don’t get faster, because there’s discoordination between all of these AI agents that are produced by these individuated efforts.”
Recruiting runs on predictable handoffs. Lee walked through the math. A sourcer surfaces a hundred candidates. Recruiters interview ten. The hiring manager meets five. Workflows are designed around those ratios. When AI spikes output at one stage and not the next, the system strains. “It will break the pipeline as a result,” Lee said. That’s the productivity paradox in a sentence.
Lagunas sees the same thing, even inside his own small firm. “We are all on hyper-individualized capability stacks now,” he said. He compared it to the familiar pain of an ATS that’s five or six years old and has survived a few talent leaders, layered with everyone’s preferences until it’s clunky and messy. “We are about to see that trend on steroids. Every recruiter has always loved to do things their own specific way, and now they really can.”
His fix isn’t heavy-handed governance. It’s shared norms. Talent ops leaders, he said, need to set guidelines that define the working style: consistent, responsible and safe, but also “repeatable, shareable.”
Lee, who spent time inside software companies, has seen this movie before. Early software development was full of “individual heroes hacking away and producing these solutions that were individually brilliant, but they were just not understandable by anybody else, and they weren’t scalable to team.” Eventually all of it had to be re-engineered to fit together. Many of the heroes left. “We’re at an idiosyncratic moment in vibe coding,” he said.
The artifact nobody opens
Lagunas brought a confession. His team asked him for a standard operating procedure for a process he runs well. He wrote a thorough one. “I can see in Google Docs, no one’s opened it,” he said.
It’s a small story about a big shift. Getting to a final document used to take collaboration: inputs, debate, revisions. The artifact was the end of that process, and the process was what made people believe in it. “Now one person can run to the finish line, and then everybody needs to pick up what I’m putting down,” Lagunas said. “My artifact is beautiful, she’s perfect, but no one’s really getting anything out of it because they didn’t touch any piece of it along the way.”
“That’s a really important insight,” Lee said. “The collective effort to create the artifact helped everyone align. That was the alignment,” he said. When one person can produce the whole thing alone, it lands on everyone else as a mandate. “And of course you get resistance, ’cause they haven’t participated in that journey.”
For TA leaders rolling out AI-built workflows, the lesson is plain. Speed of creation doesn’t equal speed of adoption. If your team didn’t help shape the tool, plan to spend time earning their buy-in.
“We can build it ourselves” has a price tag nobody sees
Both panelists agreed this may be the most expensive sentence in TA. Not because building is always wrong, but because the costs hide.
The cost you can’t forecast. Lagunas recently spoke with a head of talent acquisition whose budget is now just dollars, to be split between headcount and AI tokens. That changes the math. “We know what the current cost of these tokens are. We don’t know what the future cost of these things are,” he said. Traditional software was predictable, often locked in for three years. And when something broke, there was someone to call. “If I am putting a lot of eggs into the enterprise LLM contracts that I have, well, I am several steps away from somebody that solves the problem…”
The cost of losing ownership. Lagunas called it “the indirect expense of losing ownership of the solutions.” Lee put it more bluntly. Recruiters aren’t using particularly complicated software, he said, and “you could vibe code a viable solution that would work. But what happens when it breaks?” Or when scope creep turns a simple tool into something nobody cleaned up?
The cost of no one watching. “The IT teams are not leaning in and reviewing the things that we’re building,” Lagunas said. Few TA ops teams can even see what everyone has built. Most people frame that as a safety or compliance risk. He sees a bigger one: “processes and the way of working just completely splintering. Maybe we only care about results at the end, but that’s very shortsighted.”
The cost of one person leaving. Madhubhai flagged the single point of failure. One person builds the tool and holds all the knowledge of how it works. “And then they leave. People leave all the time, right?”
Lee added a strategic layer. Most homegrown tools today run on intelligence rented from a couple of major AI labs. That isn’t independence. “They can increase the price anytime. They can rate limit your service anytime,” he said. “There’s lots of risk in going with what I call vended intelligence.” Building truly pays off, in his view, when a company commits to owning its own intelligence through open-source or on-premise AI. Anything less is “playing around with the idea of sovereignty.”
His safer model for most companies: treat vibe-coded tools as prototypes. “Your vibe code is effectively a prototype, but the thing then gets properly rebuilt by professional engineers that know what they’re doing.”
The ATS isn’t the villain. It just isn’t everything.
Madhubhai tried to bait the panel. Have ATS vendors spent a decade convincing TA leaders that good enough is good? Is the all-in-one promise just a clever way to sell mediocre versions of everything?
Neither took it. “I think that it does a disservice to solution providers,” said Lagunas, who has worked on the vendor side. TA has always been one of the most dynamic corners of HR tech, he said, “because we love our special tools.” ATS vendors are responding to real operating pressures and resource limits. “You do need a core. Maybe you are going to choose which solutions you need best-in-breed for. But you don’t need best-of-breed for every single problem that your recruiters bring to you.”
Lee pointed to a tension inside most TA teams. Senior leaders want fewer tools. Operators want specialist ones, and they insist on them. Which side should win depends on complexity. If you hire at low volume with a simple interview process, what the ATS offers “may not be the best, but it will do.” At enterprise scale, with a complicated recruiting and interview process, “there’s no product that an ATS would natively produce that would work.” His summary: “horses for courses.”
Madhubhai sees that line from the customer side. When enterprise customers scheduling thousands of complex interviews move to an all-in-one, “I’m always like, ‘You’ll be back.’ And they have. They have come back.” A startup running 100 interviews a month is a different story. Its CFO may be right to consolidate.
That doesn’t let specialist tools off the hook. In a tighter budget climate, Lagunas said, the burden of proof is on TA. “If you think that you need to have your very special tool for a very specific purpose, bring the receipts. Bring the guaranteed ROI, not just cost avoidance.” “My recruiters are really busy” is not a business case.
Lee was less worried about consolidation itself. Tools pile up, utilization drops, buyers move on, and someone rationalizes the stack. “I think this is kind of a natural cycle,” he said. AI just makes it feel more acute, because now there’s the temptation to rebuild what you used to buy.
Where building does make sense: around the core
None of this means TA teams should stop building. Lagunas was emphatic about the upside. He’s seen teams take real pride in what they’ve made, and the social learning is unlike anything he’s watched before. “I don’t think I have seen this much cross-team, from one employer to another employer, recruiters showing each other what they’re doing, ever.” Testing whether AI can solve a specific team’s pain point, he said, “that’s just being a responsible business leader.”
The teams getting it right, in Lee’s experience, follow a clear pattern. “The TA organizations where I’ve seen to have the most success are the ones that have maintained their core systems as solutions that are robust, commercial,” he said. “But they’ll then build kind of an ecosystem of tools around that and vibe code those tools.” The sweet spot is the gap no vendor fills: workflows too niche for a market to form, or too unique to one company.
Building the core itself is another matter. Lee would only back it with a corporate strategy behind it, such as a software company testing a product before taking it to market. “I wouldn’t be comfortable with a non-product business trying to do that, to be honest.”
Scale is the deciding variable. “Vibe coding a tool makes more and more sense the smaller you are as an organization, and less and less sense the bigger you are,” Lee said. At AI-forward companies, he’s watched a familiar arc. Everyone builds. Chaos follows. Then someone notices there are 17 versions of the same note-taking tool, picks the best one, deletes the rest, and has engineering rebuild the winner properly.
What if engineering insists it can build your scheduling tool or ATS? Lagunas offered a pragmatic take. Sometimes you let them try. “Being right is not how work gets done in enterprise,” he said. “Your definition of minimum viable and my definition of minimum viable are different.” If the build comes back “fit for minimum requirements and not fit for our actual operating needs,” that’s not a failure. It’s how TA earns credibility that it knows its business.
Automation doesn’t erase manual work. It moves it.
Madhubhai put GoodTime in the hot seat. Are TA tech vendors selling real automation, or better-looking interfaces for the same manual work?
“I think it’s different manual work now,” Lagunas said. His team hands him finished artifacts: workflows, detailed candidate summaries. “I’m spending a lot more time reviewing the output and less time working through the work.” Analysis that once took him days now takes seconds. But the slow, hands-on steps used to be where thinking happened. “We’re going right to the thinking.”
Lee wanted to underline that, and he raised a point that deserves more attention. He cited his friend Martin Redstone, who asked, “What if repetitive work was actually a mental health break?” Lee remembers paging through 100 pages of candidate search results and being glad for the focus. Strip out all the routine work, and people may spend every hour on high-strain tasks. “We might end up just becoming highly stressed, burnt out,” he said. Software engineers already feel it. “They’ll tell you right now they’re doing QA.” His warning: “Job transformation that AI is triggering is leading to new jobs that may not be that enjoyable to do.”
Lagunas added a test every TA leader should apply to every tool, AI or not. “If you have implemented any tool and it has created more administrative burden at any point for your team, then you need to reconsider your solution design.” That’s not an AI problem, he said. “That has always been a tech problem.” Madhubhai added the other half: culture. No vendor, and no homegrown tool, can solve for an organization that hasn’t bought in.
So, build, buy, or ATS? How to make the call
In the rapid-fire round, Madhubhai asked each panelist where they’d put their money for interview scheduling in 2027. Lagunas, who noted GoodTime is a client, didn’t hedge: “I would say buy it.” With teams doing more manual interviewing to help combat fraud, he said, “don’t build your own scheduling.”
Lee refused to give a blanket answer. “It’s so contextual based on need,” he said. A small business may not need a specialist tool at all. But scale changes everything. “If you’re spending way too much time trying to organize a single interview, you’ve got a lot of interviews that are rescheduled, re-canceled or whatever it is, that’s where the problem starts emerging.”
The audience questions turned that into a practical playbook.
How do you know your ATS scheduling isn’t enough anymore? “You’ll feel it,” Lee said. Look for the signals: a clunky experience, hiring managers and interviewers resisting the tool, and a trend line that keeps getting worse.
How do you push back when engineering says it can build it cheaper? “Data. You need data,” said Lagunas. “Cost is only one dimension” of a real solution, alongside quality and speed. Lee questioned the premise. Who owns the tooling decision for TA? “It shouldn’t be the case that engineering can insist on this,” unless the C-suite has mandated that everything be built in-house.
How do you win a specialist tool when the company is cutting vendors? “Kill some darlings,” said Lagunas. Build credibility, bring “a really strong data story,” and know where you can give. Lee’s advice: start at the end. Define the outcome, show the pathway, and explain why the tool gets you there. “It’s just normal business case development really.”
How do you avoid regretting an AI tool purchase? Look hard at the partner. With so much change and uncertainty, Lagunas said, “the partner that you have is, I think, the one constant.” Lee agreed, then added a reality check. Every significant decision carries risk. “We only decide when the data runs out,” he said. “Get comfortable with the uncertainty.”
What’s the biggest lie TA tech tells buyers? Lagunas answered in three words: “Recruiting is broken.” Lee couldn’t top it. “This is generally how recruitment is,” he said. “It’s always gonna be messy, always gonna be tough.”
The bottom line
AI has made building cheap and fast. It hasn’t made coordination, ownership, or adoption any easier. The panel’s advice comes down to a few principles:
- Buy the core when complexity is high. Enterprise-scale, high-volume, complex processes need robust, supported interview scheduling software.
- Build around the edges. Use AI-built tools for the niche gaps no vendor serves, and treat them as prototypes.
- Count every cost. That means token pricing, maintenance, single points of failure, and who you call when something breaks.
- Protect alignment. A tool your team didn’t help shape is a tool your team may not use.
- Match the tool to your scale. What works for a 100-interview-a-month startup won’t hold for an enterprise.
Asked for one last spicy take, Lee rolled out a one-liner: “I’m pro-build ahead of buy, so long as you have your own AI in-house.” Lagunas pointed outward. Talk to peers and find out what everyone else is experiencing. “It’s gonna take a village,” he said. “So get out there.”
Weighing the decision for your own team? Read our guide, Build, Buy, or Use Your ATS? A Practical Guide for TA Leaders, to see when each path makes sense for interview scheduling.