The New Bottleneck
Modern software teams are built around four core functions - Engineering, Product, Design, and Research (EPDR) - working together to ship valuable software. Engineers code, designers visualize, researchers synthesize user behavior, and product managers align all inputs to define what users prefer and guide the team toward building the right product.
But the pace of software development is fundamentally changing. With the rise of AI tools - from IDE copilots and automated PR reviews to prompt-to-app and prompt-to-design platforms - the speed of shipping code has increased by up to 10x. The success of applications going forward transitions from technical bottlenecks to the speed and accuracy with which teams distill and act on user preferences.
User feedback tools like product analytics, user research, experimentation and prototyping are converging and evolving rapidly to match this accelerated development cycle. The following insights explore how these tools and workflows are transforming in the AI software development era.
We’re still in the early days of this new wave of tooling. As code generation continues to accelerate, teams that leverage AI to quickly iterate on user feedback will hold the strongest competitive advantage. At Gradient, we’re excited to partner with founders building the next generation of tools that make this continuous loop faster, smarter, and more impactful.
1) Direct User Research
AI agent to 10X user sourcing and interviews
Direct user research through interviews is crucial for product inspiration, allowing teams to hear and observe pain points, attitudes, and usability in users' own words and actions. Many teams also gather qualitative feedback through voice-of-customer channels like Slack, Gong, or support tickets to capture detailed user quotes, and supplement it with product surveys to collect structured insights at scale.
Market size & gap
The user research market includes surveys to moderated testing has seen major PE consolidation. Qualtrics was taken private for $12.5B in 2023 ($1.46B revenue in 2022). Thoma Bravo acquired Medallia for $6.4B and UserZoom ($80M ARR) in 2021, as well as UserTesting for $1.3B in 2022 after its $2B IPO. SurveyMonkey ($481M revenue) was acquired by STG for $1.5B in 2023. Among startups, Dovetail hit $15M ARR ($454M valuation) in 2023; Sprig reached $24M ARR ($330M valuation) in 2024. Global UX research services are worth tens of billions.
With many incumbents taken private by PEs, there're a few gaps and opportunities driven by AI in this market:
Slow and manual user interview: Recruiting the right participants is time-consuming with low acceptance rates due to unclear incentives. Scheduling delays and lengthy interviews (avg. 72 mins) add friction. Without dedicated research ops (in-house or via costly agency/ participant network), PMs and designers handle logistics and synthesis themselves, limiting research frequency and scale.
Scalable surveys lack signal quality: Tools like SurveyMonkey, Sprig, and Maze enable mass user research via survey, but questions might be biased and responses are often noisy and context-poor. Low response rates (<2%) and incentive-driven participation also dilute insight quality.
Fragmented voice-of-customer data: Teams often rely on tools like Dovetail and Condens to review research transcripts, and use platforms like UnitQ and Enterpret to analyze customer feedback from various sources - or pull user quotes directly from Slack, Gong, and support tickets. While this attitudinal feedback is rich in context, its fragmented and unstructured nature makes it difficult to synthesize into actionable product insights.
New AI Opportunity
AI persona and voice agent for interview & survey: To address scalability challenges, startups like Listen and Outset use AI personas and voice agents to automate user recruitment and conduct interviews across diverse user segments (ex: location, age, gender). Newer survey platforms like TheySaid enhance mass surveys by dynamically generating deeper, brand-aligned questions. These tools enable teams to reach 10-100x more users with faster synthesis.
Unstructured research to actionable insights: As LLMs effectively process unstructured text data, there's potential to automatically synthesize qualitative data at scale and turn it into actionable insights such as UX improvements or new feature discovery. Companies like Sauce AI are working on this approach.
2) Behavioral Analytics
AI to link session and event data at scale, enriched by qualitative customer feedback
For companies with meaningful existing users, common quantitative and behavioral user insights come through live product usage data such as individual workflow clicks or OCR-based screen recordings as session replay.
Market size & gap
Leading product analytics and session replay companies are valued at $1B-$5B, with these use cases rapidly converging. Amplitude went public in 2021 at $5B; Mixpanel and FullStory are valued at $1.03B and $1.8B respectively. These "click event native" companies all added session replay in 2024. Session replay native Posthog is valued at $920M in 2025. Recent M&A includes Contentsquare acquiring Heap ($960M valuation) and Hotjar ($40M ARR) in 2021, and Cisco buying Smartlook in 2023.
Despite use case convergence and abundant data, current tools have notable shortcomings:
Expensive metrics query and high video cost: Storing event data within a vendor's columnar database is relatively cheap and fast. However, aggregating, filtering, and querying that data into custom metrics at scale becomes expensive due to scanning massive, unindexed event logs. Session replay tools are even more costly to process and store, given the volume and complexity of video-like data formats.
Overwhelming dashboards & replays: For product analytics tools, teams accumulate countless ad hoc dashboards over time, which were created for one-off features or by former teammates and are no longer relevant. This clutters the UX with outdated views, causing user fatigue and burying relevant insights. Session replay tools offer auto-capture but lack effective aggregation into meaningful insights due to poor context-aware search, requiring teams to manually watch individual replays to learn anything actionable.
New behavior from native AI interfaces: As more dynamic UIs are built (LLM generation, voice interfaces, agent automation), interactions are beyond button presses. Teams need to understand whether agent actions succeeded, what AI content was generated, and interaction costs/ROI, since AI interactions are more expensive than traditional clicks.
New AI opportunity
AI can help bridge these gaps, and we've seen a few startups leverage these opportunities as a wedge into the market:
Warehouse-native with query optimization: Product analytics startups like Sundial have shifted from client-side SDK tracking with proprietary databases to cloud warehouse-native architectures. Rather than querying massive, unindexed event logs, this approach leverages AI to optimize queries, schemas, and summarize user behavior directly in customers' data warehouses - enabling lower costs and scalable custom analysis.
Clicks matter less, while semantic understanding rises: With the growth of AI interfaces, tools like Raindrop and Langfuse capture rich context from user-AI interactions and signals (thumbs-downs, regenerations, sentiment, etc.), store it in analytics backends, and apply LLM summarization and semantic search. This enables higher-level insights by highlighting user complaints and behavioral patterns rather than just click aggregates.
VLM to scale session replay: Startups like Human Behavior and DeepPrediction use VLMs to analyze raw replays and summarize UI behavior - identifying frustration patterns, UX blockers, and clustering similar sessions at scale. This transforms session replay from a passive, manual review tool into a proactive system that surfaces product issues and UX opportunities.
3) Controlled Experiments w/ Prototyping
AI supercharges prototyping and expands A/B testing possibilities
The best ways to accelerate product iteration are rapid prototyping and controlled experimentation. To validate prototype hypotheses and de-risk new feature launches, companies conduct experiments such as incremental rollouts via feature flagging or A/B testing to compare new features against baselines.
Market size & gap
Controlled experimentation is a $1B-$3B market, and the design prototyping market is even bigger which spans $1B-$20B. Figma went public on July 31, 2025, at $33 per share and closed at $115.50, boosting its market cap to $60B. In 2024, it generated $749M in revenue at a $12.5B valuation. Tools like Sketch (~$100M peak revenue) and InVision (valued at $1.9B) have declined or shut down. Startups like Framer reached unconfirmed $2B+ valuations with dev-focused approaches, and prompt-to-app tools are rapidly becoming the standard for prototyping.
The controlled experimentation market has growing M&A activity. Feature flagging leaders like LaunchDarkly hit a $3B valuation in 2021; Statsig reached $1.1B in 2025; and Split ($349M valuation) was acquired by Harness in 2024. In experimentation, recent acquisitions include Optimizely ($400M revenue) by Episerver and Eppo (one valued at $138M) by Datadog. UserTesting IPO'ed at $2B then sold to PE for $1.3B in 2022.
These tool markets have an opportunity to converge, as running controlled experiments to validate hypotheses is a natural next step after rapid prototyping. Current workflows bridging these tools face bottlenecks where AI could drive improvements:
Slow prototype to production transitions: Moving from prototypes to production code can take days or weeks. This delay comes from aligning with brand and design guidelines, rebuilding visuals, and the tedious conversion from static mock to production graded front-end code.
Variant explosion in experimentation: Feature flag and A/B testing tools often hit a scalability wall. Teams can test a handful of variants or flags at a time. But once adding more, the complexity explodes (e.g. 10 variants results in 2¹⁰ combinations), making comprehensive testing and analysis unscalable.
Unclear experiment drivers: Running controlled experimentation often reveals which version performs better but rarely points to why - did the layout, copy, color, or component drive new user behavior? This ambiguity makes it hard to define what the next version of detailed visual improvements should be.
New AI opportunity
Prompt-to-app/design is the new prototype: Prompt-to-app tools like Lovable, V0, and Bolt.new are gaining traction among product teams for fast prototyping in concept communication and product ideation for micro workflows. Meanwhile, prompt-to-design tools like Subframe offer detailed control over design systems and components, enabling designers/engineers to generate both hi-fi designs and production-ready code.
Scalable experimentation variants: With multimodal models and long context windows, AI can generate context-aware microcopy (e.g. CTAs, feature explanations) and UX variations (e.g. brand imagery, layout options) at scale. This enables controlled experiments with 1000x more variants, allowing teams to explore large numbers of layout and copy combinations across interfaces without manual effort. Startups like Coframe and Blok target this space.
Continuous UX optimization: Instead of running fixed experiments on human-finalized designs, we're hoping to see future tools enable real-time iteration - automatically removing poor-performing variants and introducing new designs dynamically at the component level based on live user behavior.