Lessons from 9 Tech Leaders on AI, QA and more

You’ve gotta crawl before you can walk, right? That seems to be the sentiment in modern software development teams. Organizations are being asked to move faster than ever. AI can generate code in seconds. Release cycles keep shrinking. Customer expectations keep rising. Maybe it feels more like, “Run before you can crawl.”

Here’s the catch: Speed talks, but quality comes from listening.

Across my recent conversations on the Ready, Test, Go. podcast brought to you by Applause, an interesting pattern emerged. The leaders shaping modern dev orgs are going beyond asking how to produce more; they are asking what teams should stop doing. Out with the old, in with the new, and establish better habits around it all.

Some of the biggest trends in software development and AI today focus on how teams should work. How should they gather feedback? How can they be better stewards of quality? How can they optimize and structure engineering work? How should they decide what to build — or not build?

I asked nine of our expert guests two simple, quick questions:

• What should software development organizations be doing more of?
• What should software development organizations be doing less of?

Their answers clustered around three clear areas.

1. Get closer to real users, not assumptions

It seems obvious, but software teams should spend more time with the people actually using their products. If the roadmap is solely derived around assumptions made inside the organization, that’s a critical feedback loop that opens way too late in the process — in the form of poor user adoption, negative reviews or damaged brand reputation.

Gojko Adzic, software delivery consultant and author, put it plainly. He advised teams to do more, “Speaking to actual customers and users,” and less, “Building things because one of the stakeholders thought it was a good idea.”

Internal roadmaps can quickly become self-reinforcing. When a stakeholder requests a feature, the apparatus kicks into gear. A team scopes the feature, and a sprint begins. Before you know it, a substantive amount of organizational energy has accumulated behind an idea that may never work. That’s why you need to validate with the people expected to use it.

In the episode, Adzic talked more broadly about anomalous or unexpected user behavior. These can be secret signals of potential growth — and yet they might fly in the face of all that self-imposed momentum. Rather than treat edge cases as noise, teams can treat them as evidence, just like Adzic did with his own product. The behaviors that do not fit the expected pattern can often expose where a product’s assumptions break down.

This applies even more when it comes to accessibility testing. Many people with disabilities (PWD) are excluded in planning stages. Oftentimes, the team either believes that they’ve accounted for this significant population, or they exclude it from consideration altogether.

Accessibility research with real PWD brings the impact of those assumptions into sharp focus. Jamila Evilsizor, a design and accessibility research expert, pushed for more, “Talking to users who use assistive technologies,” while doing less “Launching a product before ever talking to someone, particularly a screen reader user.”

A similar sentiment to Adzic — and there’s data to support the urgency. WebAIM’s 2026 analysis of one million popular home pages found automatically detectable WCAG failures on 95.9% of them. And it goes beyond friction in the digital experience. Applause research found that 56% of assistive-technology users surveyed regularly encountered inaccessible applications or sites since the start of 2026.

William Reuschel, Inclusive Design Practice Lead at Applause, made the same point. He called for more, “Hearing from their customers, conducting research, understanding niche use cases,” and less, “Waiting until something is launched to think about accessibility.”

Post-launch remediation is inherently reactive. Reactive quality can quickly become hair-on-fire quality. Instead, bring users into the process earlier to influence what gets designed, prioritized and tested from the beginning. No singed follicles.

The lesson is simple: if software teams want better signals, they must move closer to the people creating them. Your customers are willing to communicate. Are you willing to listen?

2. Prioritize quality, and honor people power

Traditionally, organizations treated quality as a testing phase — a notion that has been changing over the last couple decades. But, with SDLC stages becoming more fluid and collaborative, it’s an even more urgent call to action now. Testing should be treated as an operating model. Everyone should be responsible for contributing.

Lisa Crispin, consultant and co-founder of Agile Testing Fellowship, echoed this sentiment. Organizations should spend more time “Nurturing a learning culture,” she said, and less time on “Sales-driven development.”

That’s a familiar tension for almost any engineering team. Commitments are made. Deadlines tighten. Feature expectations expand. Quality takes a backseat.

The irony, of course, is that shortcuts now often lead to friction later.

Alison McGuigan, an enterprise QA director, made the point directly. “Involving QA earlier in the projects, because it's easier to challenge assumptions than it is to fix it after it's already built,” she responded, advocating for a stronger quality focus. On the flip side, teams should do away with, “Expecting that QA is just a testing round at the end, and it shouldn't be something baked into the entire process.”

Industry behavior appears to be moving in that direction. In Applause’s 2025 State of Digital Quality research, the share of respondents testing at only one stage of the software development lifecycle fell from 42% in a previous survey to 15%. More than half reported QA activity during planning, design, development and maintenance.

These results suggest teams increasingly recognize that quality is not a bolt-on, late-stage, single-threaded activity. But the operational pressure is still present. In the same research, 85% of respondents said a lack of time for sufficient testing remained a challenge.

The crunch is felt up and down the organization. Dan Vega, Developer Advocate at Broadcom, framed that tension through the craft of engineering: “I think organizations should spend more time focusing on building quality software instead of, ‘Hey, this is the most important two-week sprint of the entire year when we know it's not,’” Specially, he advocated for teams to “take time to document features, take time to write tests, take time to go through user acceptance and make sure that we're building quality software and not just getting out our 10 story points for the day.”

When it comes to what organizations should do less of, he shared a personal concern. “This could be a cop out, but I think software organizations should do a lot less layoffs,” he said. “It’s just heartbreaking. Maybe pump the brakes a little. Don’t let go of everyone yet. Let’s all figure this out together.”

A different kind of crunch here. Human expertise actually helps organizations achieve the value of AI initiatives — for those who invest in its value.

Sure, teams can generate more code faster with AI. And they can also turbo-charge weak documentation, fragile architecture and poor testing practices. Slop in the name of scale.

Atlassian’s 2026 State of Teams research found that 89% of executives say AI has increased the speed of work, yet only 6% are certain they can point to clear organization-wide ROI. Meanwhile, 87% of knowledge workers say they lack the time or capacity to coordinate across people and teams, such as on reviews, sign-offs, handoffs and alignment decisions. Everyone is focused on execution.

Thus, the productivity problem is not necessarily the typing speed of the engineering team. Sometimes they need better systems around the work they are already doing.

3. AI changes the job — the need for judgment remains

The most visible shift in engineering work today is the pivot from manually producing code to directing, evaluating and validating machine-generated output. There are clear gains to be had for those organizations that install proper processes.

Andy Sack, co-founder and co-CEO of Forum3, was emphatic that organizations should spend more time simply, “Using AI. Everyone’s talking about Claude’s software. What Claude has done for software development — between Claude and Cursor, the development stack has completely changed.” His point is that the biggest potential gains are architectural and operational.

Meanwhile, organizations should do less, “Traditional software development.” Sack sees an evolution, and cited some of technology’s leading minds in his podcast episode to support his points. “There’s a new way of software development, not just vibe coding. There’s literally a new architecture. There’s a new faster way of developing software.”

If code production itself becomes dramatically cheaper, then the structure of software development needs to change with it. It must be a tandem effort.

A randomized field study led by Microsoft and Accenture across 4,867 developers and three experiments found that access to an AI coding assistant was associated with a 26% increase in completed tasks.

Nir Valtman, CEO and co-founder of Arnica, emphasized doing more with AI. “It's never enough,” he said. On the flip side, he argued that organizations should cut back on manual coding. His broader argument is that as code generation becomes easier, engineering effort needs to shift elsewhere, like review, intent validation, security and testing.

And this is an area where developers need help. In Stack Overflow’s 2025 survey, 45% of respondents said they distrusted AI accuracy — only 3% responded that they “highly trust” it. Developers are particularly frustrated with almost-right AI solutions (66%), and 45% said debugging AI-generated code could take more time.

Rigidity toward existing, inefficient processes can be a costly outcome. Sandy Pentland, MIT professor and Stanford HAI fellow, offered the following “do less” advice: “This notion of locking things down, adding features, assuming that this is really it, and you've got to double down on what you've got now is a fundamental mistake.”

Instead, Pentland framed AI's broader potential as a means of connecting people. “[Teams] should be using AI to be able to sample opinions, to be able to see what's happening more broadly in their corporation, but also with their customers.”

Many AI initiatives simply aren’t panning out. Research from Applause’s State of Digital Quality in AI 2026 report found that 44.1% of respondents said their organizations had deactivated live AI features during the previous year because operational cost outweighed user value.

Think about how we frame the term “AI adoption.” AI experimentation and fluency is one thing. Adoption and regular use is another. By nature, there is a rush to release as vibe coding and agentic builds reach the fingertips of the masses. The theoretical has to translate to the practical. The challenge is no different for software development organizations, which need to become more disciplined about what they build and for whom.

Faster software still needs better judgment

These nine perspectives point toward a mostly coherent definition of modern software development.

Test and research with customers. Bring accessibility into the process earlier. Involve QA throughout the development process. Protect the craft of engineering. Leverage AI strategically. Build stronger systems for reviewing, validating and learning from what that acceleration produces.

The teams best positioned for the next evolution of software development trends will be the ones who make use of irreplaceable human perspective and formidable AI power. They will be the ones that get better at deciding what to build, who to listen to, what to validate and when to change course.

At Applause, we partner with innovators by extending the reach and pace of their internal teams. Our managed testing services blend human judgment with AI execution — validating software under real-world conditions, across devices and geographies, with domain experts who understand what matters most to your users and your business. Whether you're testing for accessibility, payment processing, AI safety or customer experience, Applause helps teams stay confident in what they ship while keeping pace with the speed of modern release cycles.

To hear how leading product, engineering and quality leaders are working through these decisions in real time, tune into the Ready, Test, Go. podcast brought to you by Applause. Each episode gets into the thinking behind modern digital quality strategy.

Subscribe today:

• Spotify
• Apple Podcasts
• YouTube