You have 1 article left to read this month before you need to register a free LeadDev.com account.
Estimated reading time: 5 minutes
Key takeaways:
- First impressions of AI-coding tools are almost impossible to undo. A bad early experience spreads fast and shapes adoption long after the tool has improved.
- Perception spreads faster than performance data: once a respected engineer anchors an opinion, everyone else drifts toward it.
- Don’t pitch too high, and don’t offer too many choices! Overselling sets engineers up to feel like they failed.
The latest AI model or coding agent could be getting incredible buzz online and top all the popular benchmarks, but if it gets off to a poor start, popular opinion within your engineering organization could sink it.
“The gap between perception and measurable performance is wider than most people admit. Clean performance data is rare. What you have is perception data, and it moves fast. Adoption decisions aren’t driven by it. They’re driven by stories.” Alyson van Hardenberg, engineering director at Honeycomb.io told LeadDev.
Van Hardenberg describes how a tool made its way into their product through early hands-on use rather than formal review. An engineer had been tracking a canvas library for a year and built it into a prototype, and by the time the wider team considered it, they were already heavily invested. As a result, the decision wasn’t really an evaluation anymore – it had become a choice between paying for a licence or rebuilding the system.
“No evaluation process got us there, it was the prototyper’s conviction that did,” she said.
Your inbox, upgraded.
Receive weekly engineering insights to level up your leadership approach.
First impressions and early experiences
First impressions of an AI tool – whether from a demo, initial rollout, or hearsay – impact long-term adoption and it is hard to undo, according to van Hardenberg.
“A bad early experience isn’t ‘this tool didn’t work.’ It’s ‘this tool led me somewhere wrong and I didn’t catch it.’ That sticks and it spreads through word-of-mouth. Engineers who stop reaching for a tool stop recommending it, which does the same damage as active criticism,” she explained.
Whilst there are some very enthusiastic engineers, there are many who have low trust in AI, and their views can be reinforced by social media stories of failures. “When these engineers encounter glitches or friction early on, their opinions continue to be negatively shaped, and their willingness to adopt is low,” Michael Tweed, principal software engineer at Skyscanner, explained.
“The first few weeks are critical in shaping opinions. Especially given how fast the industry is moving, if a tool isn’t positively received quickly, before long a competitor will have been updated or released and engineers will want to move,” he added.
Herd mentality
Solomon Asch’s 1951 conformity experiments show that people often align their beliefs and decisions with a majority, even when objective evidence suggests otherwise, highlighting how strongly group opinion can influence individual evaluation.
This herd mentality extends beyond controlled experiments and can also be observed in professional and technological communities.
“Peer discussions are how tools spread. Trust networks extend way beyond a company now. An engineer at Honeycomb is as influenced by what they read on Hacker News or see on Bluesky as by what their teammates say. Internal champions aren’t just up against internal skeptics. They’re up against the whole external conversation,” said Van Hardenberg.
“Once a respected engineer anchors an opinion about a tool, everyone else’s view drifts toward that anchor, even if they’ve had different experiences. I’ve watched the same people who were vocally frustrated with the pace of change in one ecosystem become confident evangelists for the latest AI release. That’s not hypocrisy. It’s a social contagion,” she added.
Managing perception vs. reality
When introducing new tools, one of the biggest challenges for leaders is managing the gap between perception and real performance. That often starts with being honest about trade-offs and resisting the temptation to oversell what new technology can deliver.
“I think being transparent about the tradeoffs is crucial – when there are so many different tools with varying levels of pricing and governance, there are times when the shiny new tool isn’t available, and as a result existing tools develop a negative reputation. You can then shift the conversation away from ‘what can’t the tool do compared to others’ to ‘how can it help’ – focusing on small use cases to build confidence and showcase success.” Tweed explained.
He stressed that AI tools shouldn’t be overhyped. Instead, teams should focus on specific problems and show how the tools solve them.
“The first few weeks are mostly about expectation-setting, and the most common mistake is pitching too high. Introduce a tool alongside a productivity narrative and engineers understand that trying it is a statement about their ambitions. When they hit friction, and they will, they feel like they did something wrong. That’s harder to recover from than honest skepticism,” explained van Hardenberg.
She added that simplifying the starting point helps adoption. “Something that stuck with me: Netflix‘s AI platform team said adoption only jumped to the broader team when they stopped offering choices. One tool, one way to start. Fewer options, more action.”
Beyond the rollout itself, leaders also have to help teams navigate the constant stream of information and hype circulating online, she added.
“The most effective thing I’ve done is a weekly AI roundup for our engineering org. Not a policy document. Just what matters this week, here’s what you can ignore, here’s what people here are actually using. The goal is to be a signal instead of more noise, and help reduce the rate of change.
Managers being transparent about their own learning curves helps too. “A director at Honeycomb shared his own Claude workflow in our team Slack last month, not as a directive, just ‘here’s what I’m trying.’ Seeing leadership learn in public helps normalize that learning,” explained van Hardenberg.
More like this
Set yourself up for success
The key lesson for leaders here is to avoid pitching a new tool too high too early.
“A leader introduces an AI tool alongside a productivity narrative, engineers understand that trying it means something about their ambitions, and when they hit friction – and they will – they feel like they did something wrong. That’s harder to recover from than skepticism. And the tool gets blamed for what was actually a sequencing problem,” Van Hardenberg said.
Maintaining a baseline level of AI maturity is essential, even as definitions of good enablement evolve over time. Doing so can help win over sceptics while reducing risk, as AI tools can quickly disrupt production systems without proper guardrails and guidance in place.
“Make sure that you are baking in the enablement effort required from day one. Not only is this important for the reasons above, but also so that other leaders in your organisation understand that these tools do require effort to be successful, and they will require continued investment over time,” Tweed said.
