The Executive Takeaway
Generative AI has pushed the marginal cost of code and basic UI toward zero. When every competitor can generate an application in an afternoon, software itself stops being a moat.
At a private executive breakfast hosted by Salient, product leaders from fintech, logistics, B2B SaaS and consumer tech met under the Chatham House Rule to tackle one question: in an AI-driven build environment, where does durable competitive advantage actually live?
The room’s answer was clear. The bottleneck in software hasn’t disappeared; it has shifted. Advantage has moved away from speed of build and toward four things: proprietary data people can trust, physical-world friction, product judgment, and the market’s capacity to absorb what gets shipped.
1. The anatomy of an AI-era moat: physicality, trust and judgment
When conversational interfaces and automated workflows can be spun up in minutes, standard SaaS features commoditise fast. One participant said the “SaaS apocalypse” is already visible: clients who once paid six figures for onboarding or workflow software are now prompting internal replacements in Claude or Cursor over a weekend.
So what actually defends a business when the barriers to execution collapse? The room identified three things.
Physical operational anchors
Software can be cloned; physical assets can’t. Leaders building in fulfilment and logistics said true stickiness lives where bits meet atoms:
“Where our stickiness comes from is the physical aspect—inventory sitting in a warehouse. Moving that isn’t something an AI agent can do; it takes an entire merchant operation deciding to shift partners.”
Proprietary, high-trust data, not just aggregation
Having data is table stakes. The moat is data that customers will stake their business on, legally and operationally. Users may turn to consumer LLMs for quick research, but once real capital, compliance or livelihoods are on the line, liability matters:
“You can’t sue an LLM if it hallucinates advice. You can sue an Australian Financial Services Licence (AFSL) holder. Trust, regulatory backing, and verified data exchanges are what keep customers paying.”
The 80/20 rule of product differentiation
Leaders converged on a simple framework. About 80% of what software does today is generic hygiene: best practices, standard dashboards and baseline UI. The remaining 20% is deep domain logic and opinionated product judgment.
The moat belongs to teams that automate the 80% and put all their human craft into the 20%.
2. The new bottleneck: discovery upstream, go-to-market downstream
The assumption that shipping faster automatically creates enterprise value has met a hard reality: throughput now outpaces human absorption. Engineering used to be the constraint. With AI, the build stage is fast, and the strain has moved to both ends of the lifecycle.
| Stage | Traditional lifecycle | The AI reality |
| Discovery | Keeps pace with build | Lagging, and often skipped |
| Engineering | The bottleneck | Rapid AI-assisted build |
| Go-to-market and adoption | Absorbs a steady release flow | The new bottleneck: sales and customer fatigue |
Upstream: foundational product discovery is collapsing
Because AI makes building feel instant, teams are shipping unvalidated features. Several leaders pointed to a gap left by the “pandemic PM era”: a cohort trained on delivery velocity rather than rigorous user discovery. They described three patterns:
- Anecdotes mistaken for evidence.Teams often treat a single customer’s story, or a synthetic AI market summary, as a real user problem.
- Multi-source synthesis replacing manual triage.Leading teams now run scrapers and agents across sales calls, support logs, CRM notes and industry forums to find blind spots before writing any code.
- Saying no is the hard skill.The hardest part of product management is no longer delivery. It is the discipline to reject features that are effortless to build but commercially irrelevant.
Downstream: customer and go-to-market exhaustion
At the other end of the pipeline, sales teams, account managers and end users are drowning.
“Our throughput is 5x, but no one is using the new features because customers are fatigued. They don’t have the time to relearn software every week.”
In response, leading teams are dropping continuous feature releases in favour of structured, seasonal “editions”, in the style of Shopify or Canva launches. Grouping enhancements into a quarterly cadence gives sales enablement and customers time to absorb change.
3. From pods to “product builders”: how AI is restructuring the org chart
The traditional agile pod, with a dedicated product manager, a product designer and a squad of full-stack engineers, is being fundamentally restructured. Leaders described these shifts already playing out:
| Dimension | Traditional model | AI-native model |
| Team topology | Separate product, design and engineering silos, with PRD handoffs | Compressed pods: one visionary or strategist, two “product builders” and an eval/QA layer |
| Role of the designer | Producing static Figma flows and UI components | Prototyping and front-end delivery: writing working front ends directly in tools like Claude or Cursor, backed by robust design systems |
| Role of the engineer | Executing tickets, fixing bugs by hand and maintaining syntax | Systems architecture and guardrails: building automation frameworks, internal platform APIs and continuous evaluation pipelines |
| The “slop” factor | Slow velocity, predictable quality gates | Maintenance debt: prototypes shipped to production with no lifecycle owner or support model |
One rule had agreement across the table: whoever builds an internal or customer-facing agent owns its long-term reliability. Leaders warned against “slop code”: internal agents built in days that slowly degrade in production because nobody was made responsible for model drift, prompt updates or incident resolution.
4. The seniority gap: why some leaders are restricting AI for junior hires
One of the most candid debates of the morning was about talent progression. If AI handles boilerplate, unit tests and syntax debugging, how does the next generation of engineers and product thinkers build intuition?
The “manual before automatic” rule
Several leaders described strict internal policies that restrict entry-level developers from using AI coding tools for their first 12 to 24 months:
“AI is an accelerator, but you have to know how to drive a manual car before you drive an automatic. If you can’t whiteboard the problem or spot bad logic without an LLM, you can’t evaluate the code the LLM writes for you.”
A thinning mid-level pipeline
Without deliberate junior apprenticeships, companies are seeing fewer engineers who can step up into high-level systems architecture.
Hiring for judgment, not just output
When hiring PMs and strategists, several leaders now ask candidates to submit their raw AI chat logs alongside their case studies. The final deck matters less than the process: the quality of the questions asked, how the candidate challenged the model’s assumptions, and how they caught its hallucinations.
5. Looking ahead: what to double down on, and what to leave behind
The roundtable closed with a clear split between the investments that will hold their value and the habits worth retiring:
| Double down on | Leave behind |
| Deep, hands-on customer discovery | Feature-factory velocity metrics |
| Rigorous evaluation (eval) pipelines | Handcrafted, unvalidated PRDs |
| Physical or workflow operational lock-in | Multi-year, rigid technical roadmaps |
| AI infrastructure and automated quality gates | Internal tools deployed without maintainers |
| Developing the “judgment layer” in PMs | Generic LLM wrappers treated as moats |
The bottom line
AI has solved the problem of building software quickly. The next era will be won by teams with the commercial restraint, deep customer empathy and structural discipline to decide what is truly worth building.
About the roundtable: this article is based on original discussion from the Salient AI × Product Roundtable series, hosted by Salient Group in Sydney. The series brings together executive product leaders for candid, off-the-record discussion under the Chatham House Rule, so comments are shared without attribution.
