The signed contract looks like a happy ending. The argument is over. The edits have stopped arriving. Someone can finally rename the file without adding another “final.” For Alex Zilberman, this is also the moment a company risks misplacing something useful: its understanding of how the deal was made.
The agreement preserves the result. The reasoning can be scattered across conversations, old drafts and the recollections of whoever negotiated it. Zilberman calls the resulting loss “corporate amnesia.” As CEO and co-founder of Chamelio, he is trying to make that experience available to the next person doing the work. A business should, in his view, get better at remembering its own decisions.
It is an unusually domestic ambition for an AI company. The problem has the feel of a household in which everybody knows the spare key exists and nobody remembers which drawer contains it. Zilberman has come to it through accounting, computer vision and fintech, bringing an operator’s interest in the small interruptions that accumulate into a working day.
A different kind of recognition
His early career included EY and PwC, where he worked as a CPA. He later became co-founder and chief operating officer of AnyVision, the computer vision company subsequently known as Oosto. He then held a partnerships leadership role at Pagaya. These were jobs concerned with building and running businesses, rather than practicing law.
The connection to legal work came partly through a colleague. Gil Banyas, now Chamelio’s co-founder and COO, had been general counsel at AnyVision. They had worked on opposite sides of the same company’s operating questions. Their next venture put that relationship inside the founding team.
Gal Lellouche brought another kind of experience. Before becoming Chamelio’s CTO, he had founded a document intelligence business acquired by Sage and held a senior AI role there. The partnership combined commercial operations, in-house legal experience and applied AI. Each founder had encountered a different part of the work that the new company hoped to connect.
Zilberman is direct about his own qualification: he is not a lawyer. That makes the composition of the team more than a biographical footnote. The legal judgment comes from people who have exercised it; his contribution includes the experience of making a company function around it.
The inbox before the platform
Legal technology was a choice the founders investigated. Zilberman describes considering several opportunities and speaking with professionals before settling on this one. Legal leaders were willing to answer cold messages on LinkedIn. Their frustration with existing tools supplied a reason to continue the conversation.
That detail matters because the origin story contains actual uncertainty. He has described legal as one of the ideas they explored, rather than a lifelong calling. The work began with finding people who would talk honestly about a problem. A reply to a cold message is modest evidence, but it is evidence from someone outside the room.
A prototype followed at the end of 2023, with ten design partners at the beginning of 2024. Its interface was email. Customers sent contracts in; redlined documents came back. For all the subsequent vocabulary around AI agents, the early transaction could be understood without a diagram.
The requests widened. Users wanted help with work beyond the first review. Zilberman’s response to customer feedback includes a firm limit: he resists reproducing an elaborate legacy workflow simply because a customer already has it. He wants to understand the intended outcome before rebuilding the steps.
“We built everything guided by customer feedback, but led by our vision.”
Alex Zilberman
A dinner with twelve legal leaders
There is a revealing dinner in the company’s early fundraising story. During a New York visit for the final investment committee pitch at Work-Bench, the founders passed on dinner with the investors. They were hosting twelve legal leaders that evening, including two existing customers.
The investors remembered it. A week later, the team was back in New York, working a legal technology conference. A monitor failed at their booth; someone from Work-Bench helped rescue the setup. Even a business selling software for the next era can find itself dependent on a functioning screen.
The anecdote offers a more concrete view of the founding team than an adjective about drive. There were two invitations competing for an evening, and the customer gathering went ahead. Existing users and prospective buyers could meet in the same room. The product would have to survive a conversation among people who understood the job.
Zilberman also puts attention into what happens after a prospect agrees to try the software. He has described mandatory pilot training and direct Slack or Teams channels with customers. The rules sound quite ordinary. Their purpose is equally ordinary: people need to learn a new tool, ask questions and receive answers while their regular work continues.
The work after the signature
In his writing on contract obligations, Zilberman returns to the period after execution. Signing an agreement creates continuing work. Someone must understand what was promised, when an obligation becomes relevant and how it connects to the business. A repository can preserve a document while leaving those questions for another person to resolve.
His argument gives the signed contract a longer working life. The information inside it should remain available when circumstances make it useful. A negotiated provision may matter months after the celebration of getting the deal finished. The end of the negotiation and the end of the responsibility occupy different dates.
That interest also explains his emphasis on institutional memory. In a March 2026 essay, he argued that collections of prompts become unreliable as policies and preferred positions change. He proposed capturing the decisions teams accept, reject and override, with controls over how that history informs future work.
The word “memory” can make this sound almost sentimental. His account is concerned with rules: versions, scope, confidence thresholds, explicit human overrides and an explanation of why a suggestion appears. A system that remembers yesterday’s judgment must also accommodate the person authorized to change it today.
Who gets to change the rules?
One practical example is a termination letter. Zilberman has described a process in which legal supplies the template and specifies which fields an agent may fill. The standard wording and formatting stay fixed. The software draws designated information from the underlying contract; the generated content still requires review.
It is a helpful example precisely because the freedom is limited. The task has a defined shape. The template has an owner. The agent has work it may perform and text it may leave alone. Responsibility remains visible in the arrangement.
He has discussed the changing legal playbook with Donovan Bell, a CLOC board member and former head of global legal operations at Intel. Their subject includes turning established positions into usable boundaries for contract decisions, while identifying matters that still require human approval.
There is a recognizable management problem beneath the technology. Writing a rule does not ensure that the next person encounters it at the right moment. Making the rule available inside the work changes its usefulness. It also makes the question of who maintains it harder to postpone.
The difficult part is agreeing to change
In March 2026, Chamelio partnered with legal operations consultancy UpLevel Ops. The arrangement paired the platform with help redesigning workflows, aligning stakeholders and managing organizational change. Zilberman identified a practical obstacle: enterprises must decide which processes to revise and communicate those decisions across the organization.
That is a familiar complication for anyone who has installed software and then discovered that the meeting about how to use it takes longer than the installation. Finance, procurement, IT and legal may share an objective while carrying different assumptions about the route.
He has also joined a discussion with Bizzabo general counsel Noa Weyl and Artificial Lawyer founder Richard Tromans about measuring legal technology’s value. The conversation questions whether activity measures, such as logins or contract counts, explain the business impact. For an operator, it is a natural concern: being busy and making progress require different evidence.
More capital, the same question
The business expanded during 2026. January’s financing announcement brought cumulative funding to $10 million. Work-Bench identifies the seed investment it led as $5.7 million. By September, Chamelio had announced a $26 million Series A led by Entrée Capital, with Work-Bench, Emerge Ventures and Bright Pixel Capital participating.
Led by Entrée Capital
Announced 22 September 2026.
At the September announcement, the company reported hundreds of customers and annual recurring revenue four times its level at the seed round. Its named customers included Wiz, monday.com, Socure and AppsFlyer. The planned use of the new money included expanding product and legal engineering teams and improving customer onboarding.
The figures mark a new scale for the company. They also enlarge the practical questions running through Zilberman’s account: what should the software learn, who approves its actions, and how does a team begin using it while the business keeps moving?
His stated ambition is a unified system of record for legal operations, one in which earlier decisions inform the next assignment. For the former computer vision operator, the object of attention has changed. There is a company’s accumulated experience to make usable, a lawyer waiting for context, and another contract arriving in the inbox.