Author: Lex Sokolin
Compiled by Jiahuan, ChainCatcher
This article explores how AI is reshaping organizational structures themselves. Companies are moving from Amazon-style "two-pizza teams" (a team of approximately 6–10 people, maintaining an agile organizational structure.) to "AI-native" teams of 3 to 5 people with significantly increased productivity.
We compared two paths:
Klarna’s AI replacement strategy ended in failure. The workforce was cut from 5,500 to 3,400, and service quality issues ultimately forced it to rehire.
Coinbase and Ramp have chosen to restructure their businesses around AI augmentation and orchestration. Coinbase has laid off 700 employees and shifted to single-person product teams and AI-powered code generation.
Ramp has built an internal AI harness used daily by 99.5% of employees, covering over 350 business skills.
In addition, we analyzed why companies like Box and Plaid have been revalued by capital markets as AI infrastructure, primarily because they control enterprise-grade data with permissions essential for AI agents to operate.
The third evolution of organizational form
Several months ago, we discussed "Zero Human Companies" and the AI economic autonomy curve:
Although forces are already pushing toward the creation of organizations with no human intervention, the current economic actors are still us humans.
The most challenging task today is transforming existing traditional companies into AI-first entities.
This is an enormous opportunity, so Anthropic is partnering with the entire private equity industry to advance it.
Beyond those impressive financial figures, we are beginning to clearly perceive another entry point of AI’s impact: the way people build and organize companies.
Organizational structure is itself a technology.
The waterfall model gave rise to hierarchical software development giants that dominated the early tech era.
The industry then shifted to lean teams using agile methodologies, and agile evolved further into the "two-pizza teams" pioneered by Amazon. This operational structure is what underpins every modern fintech company today.
But the tide has changed direction again.
Martin Harrysson and Natasha Maniar from McKinsey predicted the next version for the end of 2025:
AI-native roles essentially mean we are shifting from a "two-pizza team" structure to single-pizza teams of 3 to 5 people.
Halved in numbers, work continues as usual.
On May 5, 2026, Brian Armstrong reinforced this argument by laying off 700 people.
What did Coinbase do?
Coinbase laid off 14% of its 4,951 employees.
Part of the reason is that this is a normal market cycle operation for a company whose business and trading volume are highly correlated—its first-quarter revenue is expected to be $1.7 billion (a 26% year-over-year decline), with earnings per share (EPS) plunging 86%.
However, it is highly noteworthy how its management plans to implement AI in modern fintech/crypto companies, and their expectations for future per-capita productivity.
Coinbase engineers can now release products that previously took weeks to launch in just a few days, and this acceleration is speeding up.
Armstrong is restructuring the business lines to ensure no more than five management layers exist beneath the CEO and COO.
The pure "manager" will no longer exist—every leader must also be an individual contributor, a "player-coach" who is proficient in modern tools and capable of both leading a team and stepping in to do the work themselves.
A cross-functional "AI-native team" has fully replaced traditional teams. Coinbase has even piloted internally a single-person team that combines engineering, design, and product functions.
Coinbase, a publicly traded giant with $7 billion in revenue, is running a single-person product team.
In September 2025, Armstrong publicly stated that 40% of Coinbase’s daily code is generated by AI, with plans to increase that percentage to 50% in October.
On John Collison’s Cheeky Pint podcast, the co-founder of Stripe admitted he fired engineers who still refused to use Cursor and GitHub Copilot even after a week of enterprise license availability:
Some people just didn't use it, so they were fired.
Version V1 was a direct replacement, but it failed.
However, Coinbase is not the first fintech company to cite AI as a reason for layoffs.
Do you remember Klarna’s textbook “AI cost-cutting” experiment in 2024? At the time, it seemed to herald an astonishing surge in productivity.
But at the time, we felt this was more a tightening of the credit cycle than true innovation.
CEO Sebastian Siemiatkowski previously announced that the AI assistant powered by OpenAI handled 2.3 million conversations in its first month, accounting for two-thirds of all customer chats and accomplishing the workload equivalent to 700 full-time customer service agents.
- The total number of employees dropped sharply from 5,500 to 3,400.
- Expected profit increase: $40 million
- Customer issue resolution time reduced from 11 minutes to 2 minutes
However, all of this quickly collapsed upon encountering reality.
Customer satisfaction (CSAT) for complex tickets has plummeted, while repeat contact rates have surged.
By May 2025, Siemiatkowski admitted to Bloomberg that the company had "grown too fast." Klarna had to begin hiring again in a model similar to Uber’s—engaging students, stay-at-home parents, and workers in remote areas with flexible schedules.
The Commonwealth Bank of Australia swiftly halted 45 voice bot replacement projects within days. Taco Bell also removed voice AI from 500 drive-thru locations.
Gartner predicts that by 2027, half of the companies that have developed a "complete replacement plan" will abandon it.
Klarna's IPO still surged 30% on its first day, reaching a $20 billion valuation, which reflects to some extent that public markets are quite forgiving as long as companies correct their course in a timely manner.
But this simplistic and brute-force substitution logic—directly eliminating a human role and replacing it with a large language model (LLM)—may work for metrics focused on "quantity," but it will inevitably collapse when it comes to metrics focused on "quality."
The cost of rehiring far exceeds the initial savings. Clearly, the first attempt at AI digital transformation in the fintech sector has delivered a mixed result.
But this will certainly not be the last attempt.
Version V2 enhances capabilities, with Harness as the moat
Ramp officially launched "Glass" in early April 2026.
Seb Goddijn, an internal AI expert who worked with five colleagues to build the tool, wrote a long article. On the same day, Ramp’s CEO Eric Glyman shared it on Twitter. Within hours, the article topped the Hacker News homepage.
Goddijn sharply pointed out why Version 1 failed:
The primary barrier to AI adoption is not the models themselves, but the extreme complexity of configuring AI environments.
Glass was created by Ramp to break down this barrier:
First, automated access configuration—simply log in via Okta SSO, and all authorized internal tools (Salesforce, Gong, Notion, Linear, Snowflake, Slack, Zendesk, and Ramp’s own internal tools) are seamlessly integrated at the backend.
Second, establish the Dojo—a marketplace featuring over 350 AI skills, each represented as a Markdown file designed to teach an agent how to complete a specific task. All are stored in Git, subject to code review and version control.
An intelligent agent named Sensei will automatically recommend the five most relevant skills to new employees on their first day.
Third, build a persistent memory repository—automatically generated based on authentication connections and continuously refreshed through a 24-hour processing pipeline. As a result, the agent fully understands the employee’s team, ongoing projects, active tickets, and communication context before engaging in every conversation.
Today, 99.5% of Ramp employees use AI every day.
Half of Ramp's code is written by AI, and it's moving toward 80%. Its Chief Product Officer, Geoff Charles, has implemented an L0–L3 maturity framework, where L3 represents releasing production-grade features directly through AI agents.
Any employee still at Level 0 is effectively considered to be slacking off.
Ramp is currently valued at $32 billion, with an ARR (annual recurring revenue) of $1 billion, ranking #1 on Fast Company’s 2026 list of the Most Innovative Companies in Finance.
Klarna is trying to lower the human resource barrier through automation, while Ramp is striving to raise the productivity baseline for each employee. Coinbase lies between the two.
AI Harness
At the core of all of this is the concept of "AI Harness".
Companies like Manus have pioneered architectures that compress raw AI intelligence into repeatable business workflows, while orchestration frameworks like OpenClaw bring them to the mainstream.
A Harness is a comprehensive integration of authentication, system integration, a knowledge repository, a skills directory of team expertise, a batch scheduling program for overnight tasks, and a multi-pane interactive interface that enables analysts to run multiple tasks in parallel.
And those cutting-edge large language models are merely interchangeable components within this Harness—when OpenAI releases GPT-5.5 or Anthropic launches Opus 5, Ramp simply swaps out the model, and everything else continues to function as usual.
Anthropic's own Cowork product will be commercially available (GA) in the first quarter of 2026, featuring 11 plugins tailored for specific roles across sales, finance, legal, marketing, HR, R&D, design, and operations—the same role categorization logic as Glass's Dojo.
Once you accept that AI's productivity is shaped by business flows rather than chat interfaces, job roles naturally become the smallest natural unit of an AI organization.
This is precisely the underlying logic behind tools dedicated to building a "human-zero company" as they consider how to construct AI-first organizations. See Polsia below, followed by a rapid industry segmentation map.
The capital markets are catching up.
While many traditional software companies are struggling due to AI’s disintermediation, a certain type of player is surging逆势.
These companies built their own data moats early on and are now effortlessly layering one-time AI software on top.
Taking enterprise file storage company Box as an example: after its Q4 fiscal year 2026 earnings report, its stock price surged by 10%. Aaron Levie cut to the chase during the earnings call:
Files are, at their core, the natural units of work for AI agents.
Enterprise Advanced—the premium subscription tier featuring AI and workflow enhancements—is priced 30% to 40% higher than the traditional flagship Enterprise Plus.
Billings for the fourth quarter reached $420 million, a 5% year-over-year increase.
- Box Extract can accurately extract structured data from contracts.
- Box Shield Pro brings agentic AI directly into the access control system.
- The Professional and Extended modes of Box AI Studio enable agents to process multi-step workloads within a larger context window.
Levie remarked in an interview with GeekWire:
Except for the first 12 months after its founding, Box has never felt this much like a startup than it does today.
Did you know that up to 95% of enterprise data is unstructured? AI agents are highly dependent on this data and must be able to access it while fully preserving permission boundaries.
Whoever controls this permissioned data vault can shed the label of "cheap storage" and be revalued by capital markets as "agent infrastructure."
In the past, the market viewed Box as the slightly awkward older brother of Dropbox, with its stock price lingering around $26 for years. Today, Wall Street’s consensus target price has settled at $35.63, offering a 35% upside from the current price.
Another example is Plaid, a financial data aggregator that nearly became a subsidiary of Visa and hoped to leverage this to become a direct payment network.
But for a while, Plaid found itself in an awkward position: Web3 later rose to prominence, replacing Web2 as the new darling of financial infrastructure.
From its peak valuation of $13.4 billion in 2021, Plaid declined to $6.1 billion in its 2025 April primary market round, then rebounded to $8 billion in a secondary market tender offer in February 2026 aimed at providing liquidity to employees.
It must evolve.
Approximately 20% of Plaid's newest customers are AI-native companies—building agents that require authorized access to financial data and rely on a trusted identity foundation.
Plaid Protect’s fraud prevention platform detected 50% more fraudulent attempts than comparable identity verification tools in early 2026 testing.
Plaid Bank Intelligence, with its Retention Score and upcoming Primacy Indicators, turns customer churn prediction capabilities into a product sold back to banks.
Plaid is being repositioned as the world's largest, authorized financial transaction data corpus.
It is not a data pipeline—data pipelines have always been commodities. The real asset is the intelligence built upon it, and the proportion of AI-native customers is the strongest evidence for this argument.
A prime example is its integration with Perplexity—jointly creating a fully integrated personal finance "computer." How we miss Mint.com! (the iconic U.S. personal finance app launched in 2006)
Box and Plaid are on the same side of the same赛道.
Both were priced during the ZIRP era under the logic of "SaaS dominance," witnessed their valuations halve, and are now being repriced under an entirely new framework—the unstructured content repository and permissioned data networks as the foundational substrate that enterprises in the V2 era can be read by agents.
Version 3 is orchestration—the birth of the "one-person company"
Sam Altman has a bet with other tech CEOs on the year the first "billion-dollar solo company" will be founded.
Dario Amodei estimates the probability of occurrence within 2026 at 70% to 80%, naming three areas: proprietary trading, developer tools, and automated customer service.
Sequoia is adjusting its investment underwriting model, making "agentic leverage"—per capita income—the primary signal. In early batches of Y Combinator incubated companies, 95% of the code has been generated by AI.
In fact, some companies have already created remarkable economic leverage using AI.
In this type of company, the CEO becomes an "agent orchestrator," managing countless AI agents from a massive cockpit.
The organizational chart became a business flow diagram that can be outsourced to machines. The labor budget became a computing power budget.
The first-generation forms of such companies will operate in narrow domains—proprietary trading, developer tools, and niche consumer software with network effects—where workflows are fully digital, regulation is light, and trust costs are low.
They will be fragile, as all single points of failure are fragile.
They also struggle to penetrate regulated enterprise markets, where the name on the contract and the face are inherently structural entities.
But such companies have already emerged.
Each technological revolution destroys the roles once considered essential under the previous paradigm—“computer” (early human calculators), production line supervisors, project managers, and middle managers.
Companies that were the first to understand this new form of economic organization often gained substantial returns due to their early action.
For example, Amazon’s “two-pizza rule,” which enables it to maintain innovation even at a scale of millions of employees, is itself a moat.
Whether we ultimately end up with a "one-person company" or a "zero-human company" is not the real issue.
Today, we are still in the process of digital transformation, and delivering value across the entire economy along this trajectory will generate hundreds of billions of dollars in returns.
The real question is: Whoever can own or build the right AI harness today can design the correct organizational chart for companies in 2026.
This means upgrading this corporate superorganism so it can continue fighting and live one more day.
Hopefully, we humans can also achieve what we wish for.