Former Consensys CMO: The Evolution of Corporate Forms in the AI Era

Author: Lex Sokolin

Compiled by: Jiahua, ChainCatcher

This article explores how AI is reshaping organizational structures themselves. Companies are shifting from Amazon-style “two-pizza teams” (a team of about 6–10 people, maintaining an agile organizational structure) to “AI-native” small groups of 3 to 5 people with a major leap in productivity.

We compared two paths:

Klarna’s AI substitution strategy ended in failure. The number of employees was cut from 5,500 to 3,400, and service quality problems ultimately forced it to rehire.

Coinbase and Ramp, on the other hand, chose to restructure their businesses around AI enhancement and orchestration. Coinbase laid off 700 people, while shifting to single-product teams and AI code generation.

Ramp built an internal AI-driving framework (harness). 99.5% of employees use it every day, covering more than 350 business skills.

In addition, we also break down why companies such as Box and Plaid are being re-priced by capital markets as AI infrastructure—at the core, they control permissioned enterprise-grade data that is necessary for AI agents to operate.

The third evolution of organizational form

A few months ago, we discussed the “Zero Human Companies” and the curve toward AI economic autonomy:

Although there are forces pushing toward building organizations with no human intervention, the current economic actors are still us humans.

The most difficult task today is to transform existing traditional companies into AI-first forms.

This is an opportunity so enormous that Anthropic is partnering with the entire private equity industry to push it forward.

Beyond those astonishing financial figures, we have started to clearly sense another entry point for AI’s impact: the way people build and organize companies.

Organizational structure itself is a technology.

Waterfall development gave rise to the software behemoths of the early tech era, marked by rigid hierarchy.

Then the industry shifted to lean teams using agile methodologies, and agile subsequently evolved into Amazon’s pioneering “two-pizza teams.” It is precisely this operating model that underpins every modern fintech company today.

But the tide is turning in a different direction again.

At the end of 2025, Martin Harrysson and Natasha Maniar from McKinsey delivered their next version of predictions:

“AI-native roles essentially mean that we are moving from the ‘two-pizza structure’ to ‘single-pizza teams’ made up of 3 to 5 people.”

Halve the people—keep the work going.

On May 5, 2026, Brian Armstrong added strong evidence to this claim by laying off 700 people.

What did Coinbase do?

Coinbase reduced the headcount by 14% out of 4,951 employees.

Part of the reason is that it is still a standard market-cycle move for a company whose business is highly tied to trading volume—its first-quarter revenue is expected to be $1.7 billion (down 26% year over year), and earnings per share (EPS) has plunged 86%.

But what is especially worth close attention is how its management has planned the AI rollout path in modern fintech/crypto companies, and their expectations for future per-capita productivity.

Coinbase engineers can now release products in a few days that previously took weeks to go live, and this efficiency is accelerating.

Armstrong is restructuring business lines to ensure that under the CEO and COO there are at most five management layers.

Pure “managers” will no longer exist—every leader must also be an individual contributor; they must be proficient in modern tools and be “player-coaches” who can both lead the team and jump in personally.

Cross-functional “AI-native teams” fully replace traditional teams. Coinbase even ran an internal pilot to merge engineering, design, and product functions into a one-person team.

Coinbase, a publicly listed giant with $7 billion in revenue, is operating with single-product teams.

In September 2025, Armstrong previously stated publicly that 40% of Coinbase’s code is generated by AI every day, and it plans to raise this ratio to 50% in October.

In Stripe co-founder John Collison’s Cheeky Pint podcast, he admitted that he fired engineers who still refused to use Cursor and GitHub Copilot within a week after enterprise licenses were issued:

“Some people just don’t use it, so they got fired.”

V1 was a direct replacement—but it failed

However, Coinbase is not the first fintech company to lay off staff for AI-related reasons.

Do you remember Klarna’s textbook “AI cost-cutting” experiment in 2024? At the time, it seemed to foreshadow an astonishing productivity surge in the future.

But we believed then that this was more like a tightening credit cycle rather than true innovation.

CEO Sebastian Siemiatkowski had announced it loudly: an AI assistant powered by OpenAI handled 2.3 million conversations in its first month, accounting for two-thirds of all customer chats, delivering work equivalent to about 700 full-time customer service reps.

The total number of employees dropped from 5,500 to 3,400.

Expected profit increase: $40 million.

Customer issue resolution time fell from 11 minutes to 2 minutes.

However, once it hit real life, everything quickly collapsed.

Customer satisfaction (CSAT) for complex tickets plummeted, and the rate of repeat outreach skyrocketed.

By May 2025, Siemiatkowski admitted to Bloomberg that the company “took on more than it could chew.” Klarna had to start rehiring using a remote model similar to Uber—hiring flexible students, full-time parents, and workers in remote areas.

In just a few days, Australia’s Commonwealth Bank moved to halt 45 voice-bot replacement projects. Taco Bell also removed voice AI from 500 drive-thru restaurants.

Gartner predicts that by 2027, half of the companies that formulated “full replacement plans” will abandon those plans.

Klarna’s IPO still jumped 30% on the first day, reaching a $20 billion valuation, reflecting to a certain extent that as long as companies correct course in time, public markets are fairly forgiving.

But this simple-and-crude “replacement” logic—cutting a human job and stuffing it into a large language model (LLM)—may work for “quantity” metrics, but it will inevitably break down on “quality” metrics.

The cost of rehiring is far greater than the savings originally achieved. Clearly, the first attempt at AI digitization in fintech delivered a mixed result.

But this will never be the last attempt.

V2 is capability enhancement, with Harness as the moat

In early April 2026, Ramp officially launched “Glass.”

Internal AI expert Seb Goddijn, who built the tool together with five colleagues, published a long-form article. On the same day, Ramp’s CEO Eric Glyman retweeted it on Twitter. Within hours, the article dominated Hacker News’ homepage.

On why V1 failed, Goddijn was concise and to the point:

“The first major barrier to AI adoption is not the models themselves, but the extreme complexity of configuring the environment in which AI runs.”

Glass is exactly what Ramp built to smash this barrier:

First, automated access provisioning—after logging in via Okta SSO, every authorized internal tool (Salesforce, Gong, Notion, Linear, Snowflake, Slack, Zendesk, and Ramp’s own internal tools) is already connected at the underlying layer.

Second, establish Dojo—a marketplace containing more than 350 AI skills, where each skill is a Markdown file responsible for teaching an agent to complete a task. They are all stored in Git, undergo code review, and use version control.

An agent called Sensei will, on the first day of onboarding, automatically push the five skills most relevant to each new employee.

Third, build a persistent memory repository—automatically generated based on identity verification and continuously refreshed through a 24-hour integrated processing pipeline. As a result, whenever an agent intervenes in a conversation, it already fully understands the employee’s team, projects, active tickets, and the ongoing communication threads.

Today, 99.5% of Ramp employees use AI every day.

Half of Ramp’s code is written by AI, and it is moving toward 80%. Its Chief Product Officer Geoff Charles has rolled out an L0–L3 maturity framework, where L3 means directly using AI agents to deploy production-grade functionality.

Any employees still stuck at the L0 level are, in practice, treated as slacking.

Ramp’s current valuation is as high as $32 billion, with ARR (annual recurring revenue) of $1 billion, ranking it #1 on Fast Company 2026’s list of the most innovative companies in finance.

Klarna tries to lower staffing thresholds with automation, while Ramp is aggressively raising the output baseline for each employee. Coinbase sits somewhere in between.

AI Harness

At the core of all this is the concept of “AI Harness.”

Companies such as Manus pioneered architectures that compress raw AI intelligence and convert it into repeatable business workflows, while orchestration frameworks like OpenClaw take it mainstream.

A Harness is a comprehensive system that perfectly integrates identity verification, system integration, memory repositories, skill catalogs accumulated by teams, nightly batch scheduling procedures, and a multi-pane interaction interface that allows analysts to work in parallel with multiple lines.

And those cutting-edge large language models are merely interchangeable components inside this Harness. When OpenAI releases GPT-5.5, or when Anthropic publishes Opus 5, Ramp can simply swap out the model while everything else in the surrounding ecosystem continues to run as usual.

Anthropic’s Cowork product was officially released for general availability (GA) in Q1 2026. It includes 11 plugins targeted at specific job roles, spanning sales, finance, legal, marketing, HR, R&D, design, and operations—this job-role categorization logic is identical to Glass’s Dojo.

Once you accept that “AI productivity is shaped by business flows rather than chat windows,” job roles naturally become the smallest fundamental units of AI organizations.

This is also the underlying logic of the tools dedicated to building “zero human companies” when thinking about how to construct AI-prioritized organizations—see Polsia below, as well as the later rapid industry segmentation map.

Capital markets are catching up

While many traditional software companies are struggling because of AI-driven disintermediation, a certain type of player is surging against the trend.

These companies dug deep into their own data moats early, and now they can add one-off AI software on top seamlessly.

Take the enterprise file storage company Box as an example: after releasing its Q4 2026 earnings report, its stock jumped 10%. Aaron Levie revealed the key point during the earnings call:

“Files, at their core, are the natural work units for AI agents.”

Enterprise Advanced—Box’s premium subscription tier for AI and workflows—costs 30% to 40% more than the traditional flagship Enterprise Plus.

Fourth-quarter billings were $420 million, up 5% year over year.

Box Extract can precisely extract structured data from contracts.

Box Shield Pro introduces agentic AI directly into the access control system.

Box AI Studio’s professional mode and extended mode allow agents to handle multi-step workloads within larger context windows.

In an interview with GeekWire, Levie mused:

“Besides those first 12 months, Box has never felt more like a startup than it does today.”

It’s worth noting that up to 95% of enterprise data is unstructured. AI agents crave this data, and they must be able to access it while preserving complete permission boundaries.

Whoever controls this permissioned data vault can shake off the “cheap storage” label and be revalued by capital markets as “agent infrastructure.”

In the past, the market viewed Box as the awkward older sibling of Dropbox, and its stock price lingered around $26 for a long time. Now, Wall Street’s consensus price target is $35.63, leaving room for a 35% premium over the current price.

Another example is Plaid—this financial data aggregator had nearly been acquired by Visa and was hoping to become a direct payments network.

But for a period, Plaid was in a rather awkward position: Web3 later overtook it, replacing Web2 as the new darling of financial infrastructure.

From its valuation peak of $13.4 billion in 2021, Plaid steadily slid to $6.1 billion by April 2025 in the primary market round, and then in February 2026, it rebounded to $8 billion in a secondary market tender offer providing liquidity to employees.

It has to evolve.

About 20% of Plaid’s latest customers are AI-native companies—they are building agents that require authorized access to financial data and rely on trusted identity foundations.

In early 2026 testing, Plaid Protect’s anti-fraud platform detected 50% more fraud attempts than comparable identity verification tools.

Plaid Bank Intelligence, along with the Retention Score and the soon-to-be-launched Primacy Indicators, is selling back its customer churn prediction capabilities to banks.

Plaid is being re-priced as the world’s largest authorized financial transaction data corpus.

It is not just a data pipeline—data pipelines have always been cheap. The real asset is the intelligence built on top of it, and the share of AI-native customers strongly supports this point.

A typical case is its integration with Perplexity—co-building a complete personal finance “computer.” How we miss Mint.com! (A U.S. national personal finance accounting app launched in 2006.)

Box and Plaid are on the same side of the same track.

Both were priced in the zero-interest-rate (ZIRP) era with the logic of “SaaS kings,” watched their valuations get cut in half, and are now being underwritten again under an entirely new logic: in the V2 era, an unstructured content treasury and a permissioned data network are the foundational substrates that AI agents can read.

V3 is orchestration—“single-person companies” are born

Sam Altman and other tech CEOs are betting on what year the first “single-person company” at a $1 billion scale will arrive.

Dario Amodei set its probability for appearing within 2026 at 70% to 80%, naming three areas: proprietary trading, developer tools, and automated customer service.

Sequoia (Sequoia Capital) is adjusting its investment underwriting model, making “agentic leverage”—income per capita—the top signal. For companies from the early batches of Y Combinator, 95% of the code is already generated by AI.

In fact, some companies have already created astonishing economic leverage with AI.

In these companies, the CEO becomes an “agent orchestrator,” scheduling countless AI agents from a giant control cockpit.

Organizational charts turn into business flow diagrams that can be outsourced to machines. Labor budgets become compute budgets.

The initial form of these first-generation companies will live in narrow domains—proprietary trading, developer tools, and niche consumer software with network effects. In these scenarios, work is fully digital, regulatory burden is lighter, and trust costs are low.

They will be fragile, because every single-point-of-failure system is fragile.

They will also struggle to penetrate regulated enterprise markets, because there the name on the contract and the face attached to it are inherently structural.

But these kinds of companies already exist.

Every technological revolution destroys the roles that the previous paradigm considered critical—“computers” (early human calculators), assembly line foremen, project managers, and middle managers.

And companies that figure out the “new economic organization form” first often receive massive rewards for moving early.

For example: Amazon’s “two-pizza rule,” and its ability to maintain innovation even at the scale of hundreds of thousands of employees, in itself is a moat.

Whether we ultimately end up with “single-person companies” or “zero human companies” is not the real question.

We are still in the middle of a digital transformation process, and delivering value across the entire economy along this path will generate returns in the trillions.

The real question is: who can possess or build the right AI Harness today, and who can then design the correct organizational chart for companies in 2026?

That means upgrading this super-organism of a company so it can keep fighting—alive for another day.

Hopefully, we humans can also get what we hope for from it.

LAYOUT REFERENCE (source): total_lines=241, non_empty_lines=121, blank_lines=120