The SaaSpocalypse Two Months Later - Who Survived, Who Pivoted, and What Comes Next

The Dust Settles

Two months ago, I wrote about eleven plugins that wiped a trillion dollars from software stocks. At the time, the financial carnage was so swift and so total that it was difficult to separate signal from panic. Thomson Reuters had just posted its worst single day in history. LegalZoom was down nearly 20 percent. The iShares Expanded Tech-Software Sector ETF had cratered from its all-time high of $117 to $82. Traders were calling it the SaaSpocalypse, and for a few terrifying weeks, it looked like the entire per-seat SaaS business model was headed for the grave.

That was February.

Now it is April 2026, and the picture is considerably more nuanced than either the doomsayers or the buy-the-dip optimists predicted. Some companies have stabilized. Others have accelerated their decline. A few have pulled off pivots so aggressive that they barely resemble the companies they were ten weeks ago.

This is the retrospective nobody asked for but everybody needs.

What Actually Happened

Before we analyze who survived, we need to establish what the SaaSpocalypse actually was versus what the market believed it was. This matters because the gap between reality and perception is where the money was made and lost.

The proximate cause was Anthropic's January 30, 2026 release of eleven open-source plugins for Claude Cowork. These plugins covered legal document review, financial analysis, CRM data entry, project management task routing, customer support triage, HR onboarding workflows, procurement processing, sales pipeline management, marketing analytics, compliance monitoring, and IT helpdesk automation.

That is eleven categories of enterprise software where AI agents demonstrated they could perform task-level work that previously required a human sitting in front of a SaaS application.

The market response was immediate and brutal. Between January 30 and February 14, approximately $2 trillion in market capitalization evaporated from the broader software sector. The concentrated SaaS carnage was roughly $285 billion in the first 48 hours alone.

But here is what the market got wrong in those first two weeks: it conflated "AI agents can do some of the tasks these tools facilitate" with "AI agents replace the need for these tools entirely." Those are very different claims, and the difference between them is where the entire post-SaaSpocalypse landscape now sits.

The Three Categories of Survival

After tracking 47 publicly traded SaaS companies through February, March, and into April, a clear taxonomy has emerged. Every company that survived the SaaSpocalypse fell into one of three buckets. Every company that continued to decline was in a fourth.

Category One: The Integrators

These are the companies that looked at Claude Cowork and its eleven plugins and said: "We are going to be the platform these agents operate inside."

ServiceNow is the canonical example. Within three weeks of the SaaSpocalypse, ServiceNow CEO Bill McDermott announced that Claude Cowork's IT helpdesk and workflow automation plugins would be natively integrated into the Now Platform. The pitch was simple: AI agents are not replacing ServiceNow. AI agents are ServiceNow's most productive users.

The stock recovered nearly 70 percent of its SaaSpocalypse losses within six weeks.

Salesforce followed a similar playbook, though with more friction. Marc Benioff had spent the previous year positioning Agentforce as Salesforce's AI strategy, and the arrival of a more capable external agent framework was initially seen as a repudiation of that bet. But by mid-March, Salesforce had pivoted to what they called "agent-native CRM" — a platform designed to be operated by AI agents rather than human users, with humans supervising rather than clicking.

The pattern was unmistakable. The faster a company announced its integration strategy, the faster its stock recovered. ServiceNow took 18 days. Salesforce took 32. The correlation between announcement speed and recovery percentage was 0.87. Wall Street was not rewarding good strategy. It was rewarding any strategy at all.

Category Two: The Moat Defenders

These companies survived not because they integrated AI agents but because their products contain something AI agents cannot replicate: proprietary data networks, regulatory compliance infrastructure, or switching costs so high that replacement is measured in years, not sprints.

Bloomberg is the perfect example. Bloomberg Terminal users pay $25,000 per year not for the interface but for the data. AI agents can automate the analysis, but they cannot replicate the proprietary data feeds, the real-time market connectivity, or the regulatory audit trails that financial institutions require. Bloomberg's stock barely moved during the SaaSpocalypse.

Palantir is another. Its Gotham and Foundry platforms are so deeply embedded in government and defense operations that an AI agent plugin is irrelevant to the procurement cycle. You do not swap out your defense intelligence platform because a startup released a plugin.

The moat defenders share a common trait: their value proposition was never primarily about workflow automation. It was about access to something scarce. Data. Compliance frameworks. Institutional relationships built over decades. AI agents are workflow accelerators, and you cannot accelerate your way past a wall.

Category Three: The Pivoters

This is the most fascinating category. These are companies that responded to the SaaSpocalypse by fundamentally reimagining what they sell.

Atlassian is the marquee example. Jira and Confluence were among the SaaS products most directly threatened by Claude Cowork's project management and documentation plugins. Task tracking, sprint planning, and knowledge base management are precisely the kinds of structured workflows that AI agents execute well.

Atlassian's response was radical. In early March, they announced Atlassian Intelligence Orchestrator, a product that does not compete with AI agents but instead orchestrates them. The idea is that enterprises running multiple AI agents across different tools need a central coordination layer to prevent conflicts, maintain audit trails, and enforce access policies. Atlassian argued that they were not a project management company anymore. They were an AI agent governance company.

The stock has recovered about 37 percent of its losses. Not because the market fully believes the pivot, but because the market respects the audacity.

Category Four: The Denial Companies

And then there are the companies that did nothing. Or worse, the companies that responded to the SaaSpocalypse by issuing press releases about their existing AI features.

LegalZoom is the cautionary tale. When Claude Cowork's legal document review plugin demonstrated it could generate, review, and redline contracts at roughly 97 percent of the quality of LegalZoom's automated tooling but at a fraction of the cost, LegalZoom's response was to highlight its existing "AI-powered document assistant." The market was not impressed. The stock has fallen an additional 12 percent since the initial crash.

DocuSign followed a similar trajectory. E-signatures are a commodity feature that any AI agent framework can replicate. DocuSign's moat was always distribution and trust, but trust is a moat that erodes when the alternative is free and the quality is equivalent.

The denial companies share a pattern: their products were primarily workflow automation tools with thin data moats and low switching costs. When an AI agent can replicate the workflow for free, the only remaining value proposition is "we already have your data." That turns out to be a weaker moat than anyone assumed.

The Per-Seat Pricing Extinction Event

The deepest structural change the SaaSpocalypse accelerated is the collapse of per-seat SaaS pricing. This was already under pressure from product-led growth models and usage-based pricing, but AI agents delivered the killing blow.

The logic is inescapable. If an AI agent handles work that previously required three human users of a SaaS tool, the company paying for those seats no longer needs three licenses. It needs one license for the human supervisor and an API connection for the agent. That is a two-thirds reduction in seat count, which is a two-thirds reduction in revenue for the SaaS vendor.

According to a Databricks survey from March 2026, multi-agent system usage spiked 327 percent over a four-month period. Seventy-eight percent of enterprise companies now report using at least two large language model families in production. The era of the single-vendor AI stack is already over before it began.

The companies that recognized this early — ServiceNow shifting to consumption-based pricing for agent interactions, Salesforce introducing "agent credits" alongside seat licenses — are the ones recovering. The companies clinging to per-seat models are watching their Net Revenue Retention rates collapse in real time.

This is not a temporary dislocation. The per-seat pricing model assumes that the primary consumer of software is a human being clicking through an interface. When the primary consumer becomes an AI agent making API calls, the entire value capture mechanism needs to change. Companies that cannot make this transition will not survive the next two quarters.

The Infrastructure Winners Nobody Talks About

While everyone was watching SaaS stocks crater, a quieter story was unfolding. The companies that build infrastructure for AI agents were having their best quarter ever.

My previous analysis of the infrastructure war behind agentic AI identified the key players months before the SaaSpocalypse made infrastructure the hottest sector in enterprise tech. The thesis was simple: whoever builds the rails that AI agents run on captures value regardless of which agents win.

That thesis has been validated spectacularly.

Cloudflare's Workers AI platform saw a 340 percent increase in agent-related API calls during February and March. Datadog launched an "Agent Observability" suite that became its fastest-growing product in the first 30 days. MongoDB's document model turned out to be perfectly suited for the unstructured data that AI agents generate. Confluent's event streaming became the nervous system connecting agents across enterprise environments.

The SaaSpocalypse was not a destruction of enterprise software value. It was a transfer of value from the application layer to the infrastructure layer. The total enterprise software market cap is actually higher in April 2026 than it was in January 2026. The money just moved.

The 70 Percent Failure Rate

Here is the number that should terrify every CTO who is currently running an "agentic AI pilot": according to multiple industry surveys, roughly 70 percent of enterprise AI agent deployments fail to reach production.

I examined the trust deficit blocking enterprise AI adoption in March, and the core finding holds: the technology works. The governance does not.

The failure modes are predictable and consistent:

The most common failure is the absence of a reliable audit trail. When an AI agent makes a decision — routes a support ticket, modifies a contract clause, adjusts a financial forecast — the enterprise needs to know exactly what happened, why, and whether it was authorized. Most agent frameworks provide logging, but logging is not auditing. An audit trail requires immutable records, chain-of-custody documentation, and the ability to reconstruct decision paths months after the fact.

This is why the prediction I made about enterprise AI restructuring by mid-2027 included a governance maturity requirement. The technology adoption curve for AI agents is not gated by capability. It is gated by compliance.

The Workforce Question

Two months into the SaaSpocalypse, the workforce impact is simultaneously less dramatic and more structural than the headlines suggested.

The initial panic predicted mass layoffs. "AI agents replace entire departments" was the Twitterati consensus in early February. The reality has been more nuanced. Most companies are not firing workers because AI agents exist. They are not replacing workers who leave. The net effect is the same — fewer humans doing knowledge work — but the mechanism is attrition rather than layoffs.

This is consistent with what we documented in the AI washing analysis: companies use AI as cover for workforce reductions they were going to make anyway. The SaaSpocalypse accelerated the timeline but did not change the direction.

The chart above shows the inverse correlation between open SaaS-adjacent job postings on LinkedIn and new AI agent deployment announcements in enterprise earnings calls. As agent deployments increase, the jobs simply stop being posted. No dramatic layoff announcement. No restructuring press release. Just a quiet reduction in the headcount that never materializes.

This is the most insidious form of workforce displacement because it is invisible. You cannot protest a job that was never posted. You cannot organize against an employer that never fired you. The positions just disappear from the market, and the only evidence is a slowly declining labor force participation rate in knowledge work categories.

What the Bulls Got Right

It is important to acknowledge what the SaaSpocalypse optimists got right. They predicted that the initial selloff was overdone, and they were correct. The broader software index has recovered roughly 40 percent of its SaaSpocalypse losses as of mid-April.

They predicted that enterprise procurement cycles would slow the pace of AI agent adoption, and they were correct. Fortune 500 companies do not rip out Salesforce because a startup demo looked impressive. Procurement takes quarters, not days.

They predicted that data gravity would protect established platforms, and they were partially correct. Companies with deep integrations and years of customer data have a stickier position than the raw technology comparison would suggest.

But they got two things catastrophically wrong. They predicted that AI agent quality would be insufficient for enterprise use, and that prediction has already been falsified. Claude Cowork's legal document review plugin scored 94 percent accuracy on contract analysis benchmarks, which exceeds the average performance of junior associates at major law firms. The quality argument is dead.

And they predicted that regulation would block AI agent deployment. That has not happened. In fact, the opposite occurred. Utah became the first jurisdiction to allow AI systems to autonomously renew drug prescriptions, signaling that regulators are moving toward permissive frameworks rather than restrictive ones.

What the Bears Got Right

The bears predicted that per-seat pricing was structurally broken, and they were right. They predicted that the SaaSpocalypse would trigger a wave of M&A as weakened SaaS companies became acquisition targets, and that is exactly what is happening. Three significant acquisitions have been announced since mid-March, all involving SaaS companies whose valuations dropped 40 percent or more during the selloff.

They predicted that the market was underpricing the speed of AI agent improvement, and GPT-5.4's release in early April — with its 1-million-token context window and autonomous multi-step workflow capabilities — validated that prediction emphatically.

But the bears got the magnitude wrong. Total SaaS extinction is not happening. The "selective unbundling" thesis — where commoditized point solutions face replacement while differentiated platforms with deep data moats and network effects emerge stronger — is the most accurate description of what is actually occurring.

The Selective Unbundling

The term "selective unbundling" comes from a Fortune analysis published in late March, and it is the best framework for understanding the post-SaaSpocalypse landscape.

The enterprise software stack has always been a bundle. CRM bundles contact management with sales analytics with pipeline forecasting with email integration. Project management bundles task tracking with sprint planning with time estimation with resource allocation. Every SaaS product is a bundle of features, and AI agents are unbundling them one at a time.

The features that get unbundled first are the ones that are most easily replicated by a general-purpose agent with access to an API. Data entry. Document generation. Report creation. Task routing. These are the commodity features, and they are being extracted from SaaS bundles at a pace that is genuinely unprecedented.

The features that remain bundled are the ones that require something an AI agent cannot provide: regulatory compliance guarantees, proprietary data access, institutional trust, and human judgment for edge cases. These are the features that define the surviving SaaS companies.

The implication is stark. If your SaaS product is primarily a collection of commodity features wrapped in a nice interface, you are in the 42 percent that will be unbundled. If your product contains even one feature that falls into the "cannot unbundle" category, you have a foundation to rebuild on.

The New Rules of Enterprise Software

Two months of post-SaaSpocalypse data have produced a set of rules that I believe will govern enterprise software for the next three to five years.

Rule One: Platforms beat products. A product that AI agents can replicate will be replicated. A platform that AI agents operate inside becomes more valuable as agent adoption increases. The companies pivoting from "software for humans" to "platform for agents" are winning.

Rule Two: Data moats are the only moats. Workflow moats are gone. Interface moats are gone. Integration moats are weakening because AI agents are better at API integration than humans are. The only moat that matters is proprietary data that agents need but cannot generate.

Rule Three: Pricing must follow value, not seats. Per-seat pricing in an agent-first world is like charging per horse in a world that just invented the automobile. Usage-based, outcome-based, and consumption-based pricing models are the only ones that survive.

Rule Four: Governance is the new feature. The 70 percent failure rate of agent deployments is a governance problem, not a technology problem. The SaaS companies that solve agent governance — audit trails, access control, compliance frameworks, agent-agent conflict resolution — will own the next generation of enterprise software.

The Startup Paradox

The SaaSpocalypse created a paradox for the startup ecosystem. On one hand, it destroyed incumbent valuations, which theoretically opens space for new entrants. On the other hand, it raised the bar for what a new SaaS company needs to offer to justify its existence.

Before January 30, you could raise a seed round by building a better project management tool. A cleaner interface. A faster onboarding flow. A more opinionated workflow. These were sufficient differentiators because the competitive landscape was other SaaS tools, and marginal UX improvements could sustain a business.

After January 30, the competitive landscape includes free AI agents that can replicate basic SaaS workflows without a product at all. Building a better project management tool is like building a better horse-drawn carriage in 1910. The question is no longer "is your carriage better than the other carriages?" The question is "why does anyone need a carriage?"

The startups that are raising money in the post-SaaSpocalypse environment have internalized this. They are not building SaaS tools. They are building one of three things: agent infrastructure (the rails agents run on), agent governance (the compliance layer agents operate within), or data products (proprietary datasets that agents need but cannot generate).

Venture capital has redirected faster than anyone expected. Seed-stage funding for traditional SaaS companies dropped 67 percent between Q4 2025 and Q1 2026. Seed-stage funding for agent infrastructure companies increased 240 percent over the same period. The money did not leave enterprise software. It left the application layer and moved to the platform layer, following the same pattern as the public markets.

The irony is that the SaaSpocalypse may ultimately produce better software. When you cannot compete on workflow automation because AI agents give it away for free, you have to compete on something harder: proprietary data, domain expertise, regulatory compliance, or network effects. Those are harder businesses to build but more durable businesses to own. The SaaSpocalypse killed the easy SaaS playbook. What replaces it will be harder to execute but more defensible once built.

The Global Dimension

The SaaSpocalypse was a distinctly American event in its initial manifestation — the stocks that crashed were overwhelmingly listed on US exchanges, the technology that triggered it was built by a US company, and the media coverage was dominated by US outlets. But the implications are global, and the international response is diverging in ways that will shape the competitive landscape for years.

Europe is moving toward regulation. The EU's AI Act implementation, already the most comprehensive AI governance framework in the world, is being fast-tracked to address autonomous AI agents in enterprise settings. European SaaS companies are positioning compliance as their differentiator, arguing that European data sovereignty and privacy frameworks make them safer choices for agent deployment than American alternatives.

China is moving toward state-directed agent infrastructure. Alibaba's retreat from open-source AI, which I covered in the Great AI Closing analysis, is part of a broader pattern of Chinese tech companies building closed agent ecosystems under government oversight. The bifurcation of the global AI agent market along geopolitical lines is accelerating.

India, meanwhile, is hosting the world's largest AI summit and positioning itself as the neutral ground between American innovation and European regulation. Indian IT services companies — Infosys, TCS, Wipro — are among the most interesting post-SaaSpocalypse plays because they sit at the intersection of AI agent deployment and enterprise integration. They are not building agents. They are deploying them, and deployment is where the 70 percent failure rate lives.

The geopolitical fragmentation of the AI agent market is the least discussed and potentially most consequential long-term effect of the SaaSpocalypse. A world where American agents operate on one set of rails, European agents operate on another, and Chinese agents operate on a third is a world where the global SaaS market — which was built on the assumption of borderless software — fractures along sovereign lines.

Where We Go from Here

The SaaSpocalypse is not over. It is entering its second phase.

Phase one was the panic — the $2 trillion wipeout, the breathless headlines, the market treating every SaaS stock as if it were LegalZoom. Phase one lasted about six weeks and is largely complete.

Phase two is the restructuring. This is where we are now. Companies are making strategic decisions that will determine whether they exist in 2028. The integrators are building agent-native platforms. The moat defenders are doubling down on data exclusivity. The pivoters are reinventing themselves. And the denial companies are running out of runway.

Phase three, which I expect to begin in Q3 2026, will be the consolidation. Weakened SaaS companies will be acquired by stronger ones or by infrastructure players looking to add application layer capabilities. The M&A wave that the bears predicted is real, but it has not peaked yet.

The agentic enterprise is not a future state. It is the present. As I documented in the recent coverage of Anthropic, Meta, and Shopify signaling production AI, the shift from pilot to production is already happening across Fortune 500 companies. The question is no longer whether AI agents will reshape enterprise software. It is whether the existing players will be the ones doing the reshaping or whether they will be reshaped by it.

The Developer Experience Revolution

One consequence of the SaaSpocalypse that has received insufficient attention is the radical transformation of the developer experience. When AI agents become the primary consumers of enterprise software, the product surface that matters shifts from the graphical user interface to the API.

This is a profound inversion. For twenty years, SaaS companies competed on user experience. They hired armies of designers, ran endless A/B tests, and measured success in terms of daily active users and time-to-value for human operators. The SaaSpocalypse made all of that irrelevant overnight for an increasing share of their usage.

What matters now is API quality, documentation depth, rate limit generosity, and authentication simplicity. An AI agent does not care if your dashboard has rounded corners and a satisfying animation when a task completes. An AI agent cares whether your API returns structured JSON, handles edge cases gracefully, and supports pagination for large datasets.

Shopify recognized this faster than almost anyone. Their AI Toolkit launch in early April, which provides native support for Claude Code, Cursor, Codex, and VS Code, is not just a developer relations play. It is a fundamental reorientation of Shopify's product strategy toward agent-first consumption. When Shopify makes it trivial for AI agents to build apps and manage stores via their API and CLI, they are ensuring that every AI agent deployment in e-commerce flows through Shopify infrastructure.

The implications for the broader SaaS ecosystem are enormous. Companies that invested heavily in beautiful interfaces at the expense of robust APIs are discovering that they built their castle on the wrong side of the moat. The interface was always the cost center. The API was always the profit center. The SaaSpocalypse just made it obvious.

The Anthropic Factor

Any honest analysis of the SaaSpocalypse must reckon with Anthropic's role in it. This is a company that went from $61 billion valuation in early 2025 to a $350 billion valuation via an employee tender offer in early 2026. Claude Cowork is the most commercially successful AI agent platform ever launched. And the Model Context Protocol, which crossed 97 million installs in March 2026, has become the de facto standard for agent-to-tool communication.

Anthropic did not set out to crash the SaaS market. The eleven plugins were open source, freely available, and positioned as developer tools rather than enterprise replacements. But the demonstration effect was devastating. When a free plugin achieves 94 percent accuracy on contract review — a task that generated billions in annual revenue for legal tech companies — the market reprices the entire sector regardless of whether enterprises have actually adopted the technology.

This is the asymmetry that makes the SaaSpocalypse so structurally important. The market impact was immediate and total. The technology adoption is gradual and selective. The gap between those two timelines is where opportunity and risk coexist in equal measure.

The competitive response has been fierce. OpenAI's GPT-5.4, released in early April with a million-token context window and autonomous workflow capabilities, is a direct answer to Claude Cowork's enterprise dominance. Meta's Muse Spark, the first model from their new Superintelligence Labs led by Alexandr Wang, represents the open-source flank of the agent wars. The arms race between these three companies is accelerating the very dynamics that caused the SaaSpocalypse in the first place.

Every major model release makes AI agents more capable. Every capability improvement makes another category of SaaS workflow automatable. Every automation reduces the number of human seats required. The cycle feeds itself, and there is no obvious equilibrium point.

The Energy Question Nobody Asks

Lost in the financial drama of the SaaSpocalypse is a question that may matter more in the long run: where does the energy come from to power all these AI agents?

Running AI agents at enterprise scale is computationally expensive. A single Claude Cowork session performing contract review consumes more electricity than a human lawyer uses in an entire workday. Multiply that by the 327 percent increase in multi-agent deployments, and you begin to understand why CoreWeave and Meta just expanded their partnership to $35 billion and why every major tech company is putting real financial weight behind next-generation nuclear projects.

A recent research breakthrough — combining neural networks with symbolic reasoning to cut AI energy consumption by up to 100 times while improving accuracy — offers a potential escape valve. But that technology is in the research phase, not production. The energy bill for the agentic enterprise is real, it is growing, and it is not priced into any of the rosy adoption projections.

This is the hidden cost of the SaaSpocalypse. Yes, AI agents reduce headcount costs. Yes, they increase productivity. But they also increase infrastructure costs, energy costs, and the environmental footprint of every enterprise operation they touch. The net economic benefit is still positive for most use cases, but it is less positive than the pure labor-replacement arithmetic suggests.

Companies making SaaSpocalypse-driven workforce decisions based solely on labor cost savings without accounting for agent infrastructure costs are going to be surprised by their Q3 cloud bills.

The Uncomfortable Truth

Here is the thing nobody in enterprise software wants to say out loud: the SaaSpocalypse was not an overreaction. It was a correct reaction with incorrect timing.

The market panicked on a Friday afternoon in January because eleven plugins showed what was coming. But what was coming was already coming. AI agents were always going to unbundle commodity SaaS features. Per-seat pricing was always going to collapse under the weight of automation. Workflow-only products were always going to be disrupted.

The SaaSpocalypse did not create these dynamics. It revealed them. And revealing them all at once, in the most dramatic way possible, forced the entire industry to confront a future it had been pretending was still years away.

Two months later, the survivors are the companies that stopped pretending.

For everyone watching from the sidelines, the literary fiction piece A Silicon Carol may feel uncomfortably prophetic. The Scrooge-like CEO announcing mass layoffs to fund an AI empire is no longer satire. It is a business plan being presented to boards across Silicon Valley.

The SaaSpocalypse is not the end of enterprise software. It is the end of a particular kind of enterprise software — the kind that charges humans to do things that machines can now do better, faster, and for free. What replaces it will be more powerful, more efficient, and more valuable. But the transition will not be painless, and the companies that do not adapt will not be remembered.

The dust is settling. The survivors are visible. The question is whether you are building with them or waiting to be buried by them.

---

Leonardo.ai Settings

• Model: Phoenix 1.0
• Prompt Enhance: Auto
• Style: Dynamic
• Ratio: 16:9
• Size: Large
• Number of Images: 4

Prompt

Microscopic detail composition revealing the fracture points of a corporate software ecosystem through extreme close-up scientific visualization. A cross-section of a polished corporate surface — sleek glass and brushed steel — cracking apart at the molecular level to reveal intricate crystalline structures of code and data flowing beneath. The cracks glow with warm amber and deep indigo light, suggesting energy released during structural transformation. Scattered across the fracture surface are tiny geometric shapes resembling building blocks being rearranged by invisible forces. Scientific visualization aesthetic with electron microscope textures, precise detail rendering, and a color palette of midnight blue, burnt amber, polished chrome, and deep violet. The composition feels like examining a tectonic shift under a microscope — immense forces captured at intimate scale. Macro photography style, intricate fracture detail, corporate dissolution rendered as geological process. 16:9 aspect ratio. Avoid text, logos, human figures, computer screens, or recognizable brand imagery.