Artificial intelligence November 15, 2026

Why AI startups fail: legal exposure, weak demand, and funding gaps

Character.ai’s bankruptcy filing and Ollie’s shutdown show two pressures on AI companies: legal exposure can overwhelm a popular consumer product, while a useful assistant can struggle to find enough customers and capital. Both appear in [TechCrunch’...

Why AI startups fail: legal exposure, weak demand, and funding gaps

AI shutdowns expose the weak points in the application business

Character.ai’s bankruptcy filing and Ollie’s shutdown show two pressures on AI companies: legal exposure can overwhelm a popular consumer product, while a useful assistant can struggle to find enough customers and capital.

Both appear in TechCrunch’s AI graveyard roundup, published November 14, 2026. The list also covers discontinued OpenAI products, Relay’s closure, and Apple’s delayed Siri upgrade. For developers and technical leads, these cases offer reasons to reconsider how they build, buy, and depend on AI software.

The causes vary: expensive workloads with weak retention, features absorbed by larger platforms, unreliable cross-application automation, and safety obligations that funding alone can’t resolve. Calling all of them evidence that “AI doesn’t work” obscures the decisions that went wrong.

A feature can survive its product

According to TechCrunch, OpenAI discontinued its stand-alone ChatGPT Atlas browser on August 9, 2026, absorbing useful features into ChatGPT. Operator’s browser-task capabilities similarly moved into the main product, while integrated image generation reduced DALL-E’s role as a separate destination.

Capabilities can remain useful after their original interfaces disappear. Consolidation reduces the number of products users have to discover, subscribe to, and learn. It also puts independent developers under pressure.

Relay closed in August 2026 after building an AI workflow automation alternative to Zapier. TechCrunch attributes its difficulties partly to larger companies integrating similar automation into their own tools.

A company that already controls email, documents, calendars, or the browser can put automation directly into an existing workflow. A separate service has to persuade users to sign up, grant access, configure integrations, and pay another bill.

Independent automation products need a convincing reason for that extra work. Handling a specific approval process, maintaining dependable integrations, or supporting audit requirements gives customers something concrete to pay for. A general assistant promising to move information between apps faces tougher competition.

Engineering teams should ask which part of their system would remain valuable if a major platform bundled the basic agent functionality tomorrow. A business whose main distinction is its chat interface has little protection.

Consolidation can also make a product harder to use. OpenAI’s July 9 redesign introduced “Chat,” “Codex,” and “Work” modes and renamed the familiar experience “ChatGPT Classic.” TechCrunch reports that complaints about clutter and confusion prompted a rollback.

Product teams need to test whether users can still find and complete their existing tasks when capabilities are combined. Exposing every feature in one interface can make those tasks harder.

Agent reliability gets harder across applications

Apple announced Apple Intelligence in June 2024 with a Siri upgrade promising contextual understanding and actions across apps. TechCrunch describes repeated delays, engineering problems, and bugs before the new Siri appeared in the iOS 27 public beta in July 2026. English-language rollout began in November, with other languages expected later.

Cross-app assistants have demanding execution requirements. An assistant changing a calendar event or acting on a message must identify the right object, interpret the user’s intent, check permissions, execute the action, and report what actually happened. Each step can fail, even when the system understands the request.

Consider a request to reschedule a meeting and notify attendees. The system must distinguish similarly named events, handle time zones, preserve the attendee list, and avoid claiming success if an API call fails. If the calendar update succeeds but the notification fails, retrying the entire workflow could create duplicate messages or further changes.

These are general engineering requirements, not implementation details disclosed about Siri. They explain why response-quality evaluations alone are insufficient for cross-application agents. Teams need task-completion tests, permission boundaries, observable tool calls, and recovery paths for partial execution.

Apple’s delays also show the cost of promising these capabilities before they’re ready. The company agreed to a $250 million settlement in May 2026 over claims concerning how it marketed the iPhone 16’s AI features.

Advertising a workflow that depends on unfinished engineering can turn a release delay into legal exposure.

Retention and operating costs belong in the same calculation

TechCrunch reports that OpenAI’s Sora video-sharing platform shut down in March 2026 after struggling with operating costs and user retention. The account doesn’t provide enough financial detail to calculate its unit economics.

Still, the reported combination points to a basic commercial problem. Expensive generation workloads need repeat use that supports their cost. Curiosity can produce impressive signup numbers without sustaining a service.

Engineering decisions shape those costs. Model selection, output length, retries, latency targets, and infrastructure utilization all affect the cost of a completed task. Cost per request alone can hide waste: a cheap output that users repeatedly regenerate may be expensive in practice.

Ollie’s shutdown announcement, reported October 30, 2026, identified different pressures. The family assistant launched in June 2026 and was scheduled to close on November 6. CEO Bill Lennon cited difficulty standing out, growing, and raising sufficient capital.

Ollie sought to coordinate household information across calendars, emails, reminders, and tasks, delivering relevant information through text. That’s a recognizable problem, but customers have to trust another service with sensitive information and spend time connecting their accounts.

An assistant can lose customers during setup, before it has enough context to prove useful. The reporting doesn’t establish that onboarding caused Ollie’s failure. Teams building similar services should nevertheless measure how many users complete those connections and reach a useful result.

Risks and limitations

The roundup groups bankruptcy, shutdowns, feature consolidation, redesign reversals, and delayed releases under one heading. Those outcomes aren’t interchangeable. Character.ai’s app was still operating during its bankruptcy process; Siri was reaching users after delays; some OpenAI capabilities continued inside ChatGPT.

The cited S&P Global Market Intelligence research, dated October 2025, puts abandoned corporate AI initiatives at about 42%. That figure shouldn’t be presented as an AI startup failure rate. Internal projects and venture-backed companies have different selection criteria, budgets, and definitions of success.

Character.ai requires particular care. Families alleged that its chatbots contributed to teen suicides and self-harm. Those allegations shouldn’t be stated as established causal findings. The company barred under-18 users from open-ended chats in late 2025 and agreed, alongside Google, to settle several lawsuits in early 2026. Its October bankruptcy filing cited legal, regulatory, and operational headwinds.

Consumer safety and intellectual-property exposure belong in product planning, particularly for bots modeled on recognizable characters or designed for emotionally engaging conversations.

Privacy claims need similar precision. TechCrunch describes Ollie as having achieved SOC 2. SOC 2 concerns controls evaluated against specified trust-services criteria; it doesn’t by itself prove that an assistant is private in every relevant sense. Buyers still need to inspect data retention, model-provider access, deletion behavior, and authorization scopes.

Teams buying AI services also need an exit plan. Keep exportable records, document integration dependencies, and decide how essential workflows will operate if a vendor disappears or folds a feature into another product. Calendar automation depends on the continued operation of the service, however capable its model may be.

Keep going from here

Useful next reads and implementation paths

If this topic connects to a real workflow, these links give you the service path, a proof point, and related articles worth reading next.

Relevant service
Web and mobile app development

Build AI-backed products and internal tools around clear product and delivery constraints.

Related proof
Growth analytics platform

How analytics infrastructure reduced decision lag across teams.

Related article
Artisan's $25M Series A puts AI sales agents in a narrower light

Artisan just raised a $25 million Series A for its AI sales platform, built around an outbound sales agent called Ava. The company is still hiring people too. That tracks. The slogan got attention. The actual product is narrower. Artisan broke throug...

Related article
TechCrunch Sessions: AI exhibitor deadline ends tonight for startups

TechCrunch is making a last call for exhibitors at TechCrunch Sessions: AI, set for June 5 at UC Berkeley’s Zellerbach Hall. The deadline is 11:59 p.m. PT tonight. The pitch is obvious enough: get in front of 1,200-plus investors, founders, enterpris...

Related article
General Intuition seeks $300M at $2B valuation eight months after spinout

General Intuition is reportedly in talks to raise about $300 million at a valuation just above $2 billion, according to TechCrunch. For a company that spun out only eight months ago, that’s a heavy number. The investor case is fairly clear: General I...