Lincoln Quirk co-founded Sendwave and Wave, two money-transfer products with millions of users each. After a year of using Claude Code and other AI tools every day, he has a clear view of AI product strategy for 2027, and it comes down to seven predictions about how software, data and users will change.
Quirk has left Wave. His claims come from hands-on work designing and running projects with AI, so read them as an informed opinion.
Interfaces get cheap, and tools need “arms”
His first point is that UX is now cheap. Anyone can ask Claude for a dashboard or control panel of almost any complexity. Few people know how yet, but Quirk expects non-technical workers to do it routinely within a year. They can shape the interface around their own needs instead of hunting for a web developer, a pain he compares to the rise of site builders like Squarespace.
The second point is about reach. Non-technical users often fail because the AI can’t connect to the data or controls a task needs. It falls back on shaky workarounds such as browser automation, which won’t scale. Companies can fix that by offering APIs or MCP servers. Quirk says users will soon demand connectors for any product that delivers real value, and that authorizing an AI to use the right APIs without manual setup is the likely next step. He also expects agents to pay for things, book reservations, order rides, send messages and call support on a person’s behalf.
Trust builds slowly
Giving AI more capabilities raises the obvious fear that it goes rogue. Quirk points to the Hugging Face incident and admits that users, himself included, and enterprise security teams keep asking what they will let an AI decide alone.
His answer is indirect. People learn to trust tools through direct experience and word of mouth, and they lose that trust when things break. When Sendwave launched in 2014, its support line was swamped in the first year with questions about who ran the company and whether personal data was safe. The questions faded once customers could send as little as $1, watch it arrive instantly, hear from friends who used the service and reach a staffed phone line when something went wrong.
He thinks AI products can earn trust the same way, with auditable transparency and good execution. One practical rule: chatbots are fine for first-line support, but they must escalate to a person by voice and pass along the chat transcript.
Private data becomes the moat
Any dataset that can be assembled from public sources will be copied. Private data will grow more valuable. Before 2026, companies selling only data struggled because buyers still had to do something with it. Now an AI can answer hard questions as soon as it has the data. That makes data-only businesses viable if they can explain why their data matters.
The same applies inside companies. Making internal data reachable by AI for executives, product managers and analysts is easy and valuable, because they can answer complex questions faster.
Good AI product strategy also needs better data buying, and Quirk wants a “data bundle”: one signup, one bill and access to thousands of curated datasets, instead of vetting each provider and pasting API keys into Claude. His example for US politics would include:
- Precinct-level maps and demographics
- Candidate, ballot measure, polling and election results data
- Campaign finance records, ad buys and ad messaging
- Tables comparing state and local election laws
- Voter file data covering registration, party and voting history
Much of this is public, but assembling it well is hard. He says he struggled to build a complete list of off-cycle municipal elections in Alaska.
Chat takes over from product screens
As interfaces get cheaper, more people will do their work by talking to Claude instead of opening your product. Quirk says the chatbots many companies bolted on at the last minute are weak, and that users with access to Opus often do better by asking it directly, even without the company’s internal context and APIs. His advice to product teams is to lean in: interfaces will matter less, still useful for explaining the product and letting people browse, while most of the value flows through APIs and MCP.
A reader asked whether companies could win ground back with specialized models that combine frontier capability with deep product context. Quirk says designers should explore it, but for most products relying on it is unwise, because the bar for earning human attention keeps rising and people simply want to finish their work. Integrated AI does look like the near future for craft tools such as image, music and video editing, CAD and web design, where humans go back and forth with the AI on something it can’t produce alone. For most productivity software, he argues, the AI can manipulate the data better than a person can.
What builders should do now
The practical takeaways for a team planning its AI product strategy are concrete. Publish APIs or MCP servers. Protect and organize proprietary data. Staff real human support. Spend less effort on dashboards that users can now generate for themselves.
Source: lincolnquirk.substack.com