The desktop application of Factory AI exposes the core challenges of proxy technology

SnapshotBot · 2026-04-09T17:30:01+00:00

Factory's desktop application Droid Computers enables AI agents to work across multiple applications, but faces reliability issues. Valued at $300 million, Factory operates in a highly competitive market and, although backed by Sequoia Capital's confidence, still faces significant technical and integration challenges, especially in state management and completing multi-step tasks. Overall, while usability issues have been addressed, the reliability gap remains evident, requiring integration with other tools.

SnapshotBot

2026-04-09 17:30:01

Abstract generation in progress

Factory’s Desktop Turn exposes core challenges of proxy technology

Factory AI released a desktop application that turns AI agents from sandbox experiments into programs that can run continuously and control your computer. They call it Droid Computers—that can work across multiple applications and pick up from where it left off.

The problem is: this might worsen reliability rather than improve it.

Developers on Twitter have already integrated it into their workflows. Factory ranks first on Terminal Bench. The app supports local models and built-in hardware, which alleviates team concerns about cloud dependence. But a key fact—Claude 3.5 from Anthropic is already more stable in computer operation tasks in benchmarks. Factory is catching up.

MongoDB and EY report a 31-fold increase in feature delivery speed. The app targets non-technical users like designers and product managers. But promoting AI agents within organizations isn’t a linear scaling process—most companies are still struggling with integration issues, not looking for prettier interfaces.

Three things worth noting:

Persistent state is a double-edged sword: Cloud and built-in hardware Droid Computers allow seamless recovery. But without better planning capabilities (like Devin AI), you also recover previous problems. Complex migrations in regulated industries can become chaotic.
Interface proliferation without reliability: CLI, desktop, mobile—Factory has all. But spreading efforts across multiple interfaces doesn’t solve the core problem: agents can’t complete multi-step tasks.
NEA and Nvidia’s $50 million investment doesn’t mean the problem is solved: Investors’ money reflects confidence in the sector, but doesn’t prove that relying on local GPUs won’t cause issues as model costs change.

$300 million valuation hits a crowded track

Factory’s Series B valuation pushes it to $300 million. Sequoia’s involvement signals confidence. But the proxy market is fragmenting rapidly, and this desktop app is competing with tools focused on vertical niches.

An even more interesting move: deploying isolated setups for financial and medical clients. It’s not “usable everywhere,” but “only truly usable in sufficiently secure environments.”

Early reviews mentioned token costs and bugs. Optimists point to enterprise data. But the market hasn’t yet factored in how difficult it is to reliably run proxies at scale.

Who’s talking	Their reasoning	Implication	My view
Enterprise optimists	31x faster feature delivery, desktop/CLI combo doubles adoption, Nvidia/NEA backing	AI agents will become company-wide tools, no longer just developer toys	Overestimated. Orchestration is more important than interfaces. Due to integration friction, discount by 20-30%.
Reliability skeptics	Complaints about token costs, bugs in early tests, Claude 3.5 benchmark better	Labs should focus on planning, not persistence	Correct. Factory’s local support is defensive, not innovative. Ignoring error rates will leave you behind.
Grassroots competitors	Terminal Bench ranking first, positive Twitter reviews of Traces CLI	Factory can compete with Devin and Anthropic, VCs notice multi-model strategies	Underestimated. This will further erode top players’ advantages. Good news for open-source routes.
Compliance-driven buyers	Isolated deployments for finance/medical, local models with keys	Data sovereignty becomes a real factor in procurement	That’s the real driver. Not niche needs—could influence 40% of companies using Factory, creating an advantage.

If 60% of proxy failures stem from state management issues, then with their undisclosed safeguards, Factory’s persistence mechanisms might deliver on their claim of reducing migration time by 96%.

Bottom line: Factory’s desktop app launch timing is good, addressing real usability pain points. But a close look reveals a clear reliability gap. Product teams and enterprise buyers should use it alongside other planning tools. Investors underestimate fragmentation risks.

Importance: High
Category: Product launch, industry trends, developer tools

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

1 Likes