From idea to impact – Panel discussion | Craft 2026

June 04, 2026

From idea to impact – Panel discussion (Panel Outline)

At Craft 2026, moderator Matty hosts a preview of October’s Compass AI & Tech Summit with leadership, AI, data, UX/UI, and product track representatives. Rather than deeply answering the event theme, “From idea to impact,” the cross-disciplinary panel explores what AI has changed, where acceleration helps or hurts, how professional boundaries and employment may shift, what questions remain open, and two audience questions about misconceptions and hype.

1. Panel and conference framing

1.1 Matty — moderator

Matty is a former member of the Stretch conference team.
He frames this session as a preview of how and why the Compass summit will address AI-era product building in October.

1.2 Emese Pogácsás — leadership

Pogácsás is a former VP of engineering and current leadership coach.
She curates a leadership track balancing philosophical and practical organizational challenges.

1.3 Richard Roman — AI

Roman is CTO of the small AI education startup New Technology.
As an “AI fanboy/nerd,” he wants deep technical sessions, including efficient GPU utilization.

1.4 Tamás Németh — data engineering

Németh works on the DataHub metadata platform and previously led data engineering at Prezi.
His track prioritizes hands-on talks, war stories, and lessons speakers do not normally disclose.

1.5 Zsuzsa — UX/UI

Zsuzsa represents the former Amuse conference, now the UX/UI track.
Its scope includes research, design, culture, leadership, and AI’s effect on both researchers and designers.

1.6 István Szabó — product

Szabó is a product manager at SAP Emarsys and a four-year product-track organizer.
The track asks what product management means with AI, how development changes, and what makes an AI-era product compelling.

2. What AI has genuinely changed

2.1 AI becomes an everyday subject

Matty recalls adding “AI” to topics for attention one or two years earlier.
By early 2026, subway passengers discussed ChatGPT schoolwork and Gemini cookbook organization.
The question is no longer whether AI matters but which effects survive the noise.

2.2 Trusted data becomes foundational

Németh contrasts analysts’ curated dashboards with agents directly answering questions and generating charts.
Autonomous answers are only trustworthy when organizational data is clean and governed.
AI makes previously hidden data-quality debt impossible to ignore.

2.3 Pogácsás’s K-shaped economy

Companies and people using AI effectively move upward while those who do not move downward.
Hype can obscure this widening separation.
Conference attendees are already better positioned because they are actively learning.

2.4 The unresolved economics of AI products

Szabó asks how AI products can be priced against expensive, potentially unviable operating models.
Adoption depends on who can afford the technology and for how long.
A planned Compass talk by Vera will address AI pricing.

2.5 Information without wisdom

Pogácsás says society is “drowning in information but starving for wisdom.”
AI can instantly generate ten plausible, contradictory options.
Judgment and clarity—not option generation—determine which path creates value.

2.6 UX skepticism and research acceleration

Designers fear automatically generated interfaces will worsen user experience.
Zsuzsa sees strong value in rapid prototyping.
She rejects synthetic users and replacing human interviews with agents.
AI can greatly accelerate research analysis and synthesis, delivering real-user insight while it is still useful.

3. The reality of acceleration

3.1 Prototypes shrink from weeks to hours

Szabó says engineering prototypes that once took two weeks can take hours or minutes.
Earlier user exposure shortens the learning loop.
Faster construction does not automatically produce an organization capable of using the learning.

3.2 Individual output can create institutional chaos

Szabó invokes Gergely’s keynote distinction between individual productivity and institutional improvement.
Individuals generating more in slightly divergent directions create exponentially more coordination noise.
The organization needs alignment, not just faster contributors.

3.3 AI slop requires quality controls

Roman objects that information overload follows from poor AI use rather than AI itself.
New users can generate enormous specifications that colleagues cannot use.
Telling recipients to use another AI to comprehend low-quality output does not repair it.
Controls and quality guarantees may become a new professional specialty.

3.4 Data democratization exposes old debt

Product and business staff can now prompt data tools to create dashboards themselves.
Agents lack analysts’ tacit knowledge about which datasets and filters are safe.
AI does not necessarily slow work; it reveals data problems organizations previously swept under the rug.

3.5 Value is harder to measure than production

Leaders can count lines of code, bytes, dashboards, and features.
Those output measures do not say whether customer or business value increased.
AI makes teams run faster—including in the wrong direction.
Clean data and rapid user research help select the right direction.

4. Blurring professional boundaries

4.1 Two people can now build a product

Matty imagines one product/UX expert and one technologist producing in days what once needed a larger team.
The overlap between product, design, and engineering is clearly expanding.
The prospect is both empowering and threatening.

4.2 Build to learn is not build to earn

Szabó cites Marty Cagan’s distinction.
A product manager can generate a prototype to test desirability or feasibility.
Production software serving millions must still be scalable, reliable, traceable, manageable, and maintainable.
Producing a UI in five minutes does not confer design expertise; generating code does not confer engineering expertise.

4.3 Unknown domains hide bad output

People can evaluate ChatGPT more safely in fields they already understand.
A novice may accept a poor Lovable interface because they cannot see missing UX reasoning.
A non-engineer may ship code that leaks personal data because they cannot recognize security defects.
AI makes senior experts faster while increasing the importance of domain knowledge and critical thought.

4.4 Would you bet money on the answer?

Németh recalls feature-usage graphs that excited managers until analysts found broken data.
A plausible agent answer is not enough for an all-in product decision.
Experience is what prompts the challenge: would you risk your own money on it?

5. Will AI replace people?

5.1 Fewer people versus more ambition

One view says a person doing four people’s work will cause three positions to disappear.
Roman expects fewer, more effective workers.
The counterproposal is to retain four people and build much more rather than exchange salaries for a huge AI bill.

5.2 Smaller teams concentrate knowledge risk

Losing one person from 20 removes roughly one-twentieth of organizational knowledge.
Losing one of three engineers operating complex agents may remove one-third.
Codified agent knowledge in markdown may mitigate the risk, but the panel does not treat that as settled.

5.3 AI amplifies unhealthy product-engineering relations

A nontechnical PM may prefer an agreeable agent that never calls an idea impossible.
That PM can hand engineers a prototype and demand productionization in two days, ignoring scale and security.
In an organization already divided by mistrust, AI worsens conflict and breaks delivery.

5.4 AI amplifies healthy collaboration

Product and engineering partners can use prototypes as shared thinking objects.
Teams already able to explain risks and learn across disciplines gain tremendous leverage.
Matty argues AI may replace PMs unwilling to learn technology and engineers unwilling to communicate it.

5.5 Industrial revolutions remove tasks and create work

Computers and automated call centers changed employment over years, not overnight.
The present learning curve is faster, so workers should actively learn rather than seek reassurance.
Pogácsás distinguishes welcome task removal from job elimination; AI has so far given her more work.
Matty cites a Guardian analysis of 200 years in which technology created more work than it destroyed.

5.6 Traffic lights as the historical example

Police officers once manually directed traffic.
Automated signals removed that task but created manufacturing and infrastructure work.
Accidents then created a need for specialized traffic investigators drawing on former officers’ knowledge.
New systems reconfigure expertise rather than merely deleting it.

6. Questions panelists want Compass to explore

6.1 Roman: usable agent harnesses

Roman wants a harness he can download and start rather than painfully invent through failure.
He also worries October’s chosen speakers and subjects may be obsolete by October.

6.2 Németh: hiring and planning at higher throughput

Interviews must test different skills when people no longer work primarily through an IDE.
His company shipped a quarterly plan in one month, forcing product planning and roadmaps to accelerate too.

6.3 Pogácsás: inspiration rather than answers

She wants honest accounts of what worked and failed.
Attendees should translate those ideas into their own contexts rather than expect universal prescriptions.

6.4 Szabó: organizational productivity and prioritization

Bushra’s planned talk covers successes and failures in organization-wide AI transformation.
Damian Stooke will address discovery and prioritization when almost anything can be built.
Cheap construction makes deciding what deserves construction more important.

6.5 Zsuzsa: evidence from real UX practice

She wants examples of how companies use AI for prototypes, design, and research.
The purpose is to improve professional practice without outsourcing judgment or human understanding.

7. Audience Q&A

7.1 Q1 — Biggest misconception about building AI products

The panel distinguishes building an AI product from using AI while building a product.
Szabó warns against adding a chatbot or assistant merely because competitors do.
Gergő Horányi of Wise will present “No One Told You Need Another AI Assistant.”
A sound AI strategy solves a real customer problem perhaps ten times better than before.
The misconception is that an AI feature automatically constitutes customer value.

7.2 Q2 — Which AI trend is overhyped?

Roman says none: AI is underhyped, fast-moving, frightening, and exciting, though he jokingly retracts “Skynet.”
Another view calls AI-generated art and music overhyped.
Roman argues people will still pay for visible human effort and performance.
For functional party music, he does not care whether a machine generated it.
Zsuzsa finds AI music in films noticeably unpleasant but labels that reaction subjective.

8. People & References Cited

8.1 Panelists and speakers

Matty — moderator and Stretch alumnus.
Emese Pogácsás — leadership-track curator.
Richard Roman — AI-track representative.
Tamás Németh — data-engineering representative.
Zsuzsa — UX/UI representative.
István Szabó — product-track representative.
Gergely — keynote speaker cited on institutional productivity.
Marty Cagan — source of “build to learn” versus “build to earn.”
Vera, Bushra, Damian Stooke, and Gergő Horányi — planned Compass speakers cited by panelists.

8.2 Organizations, tools, and sources

Compass AI & Tech Summit — October event previewed by the panel.
New Technology, DataHub, Prezi, SAP Emarsys, Wise — panelist or cited-speaker organizations.
ChatGPT, Gemini, Claude Code, Lovable — AI tools mentioned in examples.
The Guardian — cited for a long-run study of technology and employment.
K-shaped economy, AI slop, agent harnesses, synthetic users — recurring concepts in the discussion.

Video: https://www.youtube.com/watch?v=e-7ONBF_Z_Q — Transcript via yt-transcript.sh; outline generated from the transcript.