Whether we’re ideating on a company we want to build or shaping an investment thesis, it starts the same way: a point of view on what’s next. That methodology is what drives success by design.
This year, we asked our partners across cybersecurity, software infrastructure, fintech, digital health and GTM to share the shifts they’re seeing and the opportunities they’re most excited about.
Cybersecurity
Cybersecurity Accelerates: From Exposure Management to an AI Arms Race
Cybersecurity is entering a new phase defined by speed. The traditional exposure management model of scan, prioritize, remediate rests on a critical assumption that defenders have time. AI is dismantling that assumption. Frontier models already demonstrate the ability to discover zero days and assemble sophisticated attack chains that once required elite human expertise. As these capabilities spread, advanced attacks are in the process of becoming commoditized by AI, pushing time to exploit toward zero. The result is not just more risk, but a collapse of the defender’s temporal advantage. Managing exposure after the fact becomes irrelevant when vulnerabilities are discovered and exploited faster than humans or traditional tools can respond.
At the same time, AI unlocks a countervailing force for defenders: true secure by design. While attackers accelerate downstream, defenders can finally move upstream. AI makes it economically and technically viable to eliminate entire classes of weaknesses before systems are deployed by enforcing secure coding practices, refactoring legacy software, and hardening infrastructure by default. This shift extends beyond application code to the foundations of the enterprise stack. Identity is the clearest example. Despite being the dominant attack surface, most enterprise applications still lack native integration with centralized identity providers due to legacy complexity and cost. AI makes it possible, for the first time, to automatically retrofit identity into existing and homegrown applications, enforce least privilege by default, and remove a massive class of identity driven exposure. As secure by design takes hold, many of today’s commodity findings simply disappear, shrinking the surface area that exposure management tools were built to monitor in the first place.
The implication is stark: exposure management as a concept is breaking down. The scan -> prioritize -> remediate loop assumes a world of static systems, known vulnerabilities, and delayed exploitation. That world is ending. What replaces it is a continuous, real time competition between AI systems, offensive and defensive, operating at inference speed. Defenders will no longer manage lists of issues. They will manage adversarial capability.
Security teams will reason in graphs rather than inventories, focusing on attack paths, emergent weaknesses, and how their defensive AI compares to the attacker’s. This shift will give rise to new categories, including autonomous penetration testing, continuous attack path analysis, and real time zero day discovery and mitigation. In the next era of cybersecurity, the question will not be what exposures exist, but whose AI sees the problem first.
Amir Zilberstein
Amir Zilberstein is a Managing Partner at Team8, where he builds and invests in Cyber and Software Infra businesses.
Resilient Purpose-Bound Infrastructure
Infrastructure now assembles and changes at machine speed. In this world, resilience can’t depend on human review, static policies, or postmortems. It has to exist at design and runtime. Systems must understand why they exist and continuously constrain themselves to that purpose—otherwise automation simply accelerates fragility.
The real gap in modern infrastructure isn’t visibility, controls, or data. It’s context. We know what is deployed and what is allowed, but not whether it makes sense for the system’s mission. As AI increasingly builds and modifies cloud, identity, and application environments, configurations without purpose become exposure. Complexity turns into attack surface.
Purpose-bound infrastructure introduces an intent-aware control plane. AI learns how an environment is meant to operate—its services, data flows, identities, and standards—and enforces only what serves that intent. Permissions tighten, drift self-corrects, and failures or attacks are constrained by default. Resilience stops being reactive and becomes ambient: not enforced by process, but by purpose.
Liran Grinberg
Liran Grinberg is the Co-founder and Managing Partner of Team8, where he invests in Cyber and Software Infra companies.
Software Infrastructure
The Gap Between AI Expectation in the Enterprise and Reality Needs to Be Bridged with the Right AI Infrastructure
In 2026, the ‘AI Summer’ ends and the ‘Operational Reality’ begins. We are moving from the era of ‘Model IQ’ to ‘System Reliability ‘ as part of the AI infrastructure. At Team8, we are focused on two defining layers of the 2026 stack:
One, the challenge for agents has shifted from reasoning to Contextual Integrity. The winners will be the “Semantic Glue” that transforms fragmented systems of record into a unified, agent-readable truth, enabling autonomous action that honors business logic, not just data schemas.
Two, as AI systems move from demos to mission-critical workflows, AI evaluation becomes the bottleneck to real adoption. In 2026, the question won’t be can models reason, but whether agents are reliable, predictable, and aligned with business intent at scale. AI Eval is the missing layer that turns probabilistic systems into operational ones, measuring not just accuracy, but behavior, consistency, drift, and failure modes across real world contexts. As enterprises deploy agents that act across data, tools, and systems of record, evaluation becomes the control plane that makes trust, governance, and iteration possible. That’s why AI Eval isn’t tooling, it’s infrastructure, and one of the defining ideas of the next wave of AI.
Aviad Harell
Aviad Harell is a Managing Partner at Team8. He builds and invests in Cyber and Software Infrastructure companies.
Fintech
From Ambition to Execution: Closing the AI Gap
One thing that stood out for me as we focus on the value of AI is the MIT study showing that only 5% of companies have seen real returns from their AI projects. It’s a striking number, but not a surprising one. At Team8, we’re constantly engaging with founders and enterprise leaders navigating this exact challenge. AI might be embedded in strategy decks, but in most cases, it’s not embedded in the business. Too many initiatives launch without foundations: clean data, long-term capital, aligned talent, or clear incentives. And proof-of-concepts don’t scale themselves. Real value only comes when AI is operationalized, when it flows through workflows, decisions, and culture.
What separates the 5% who are currently succeeding isn’t just technology, it’s also about leadership. The most important shift we’re seeing is that transformation starts at the top. Raising that 5% figure is critical. The institutions that succeed won’t just have the best models, they’ll be the ones that made AI executable. And when we look at the companies that break through, their success will most likely come down to two things: the strength of their technology, and the clarity of leadership driving it.
Rakefet Russak Aminoach
Rakefet Russak-Aminoach is a Managing Partner at Team8, where she builds and invests in Fintech companies that are poised to redefine the financial services landscape.
Agentic Payments Takes Both Consumer and B2B Turns
Agentic payments will manifest differently across consumer and B2B sectors. In the consumer space, AI agents integrated into smart devices and personal finance apps will autonomously manage subscriptions, optimize utility payments based on real-time pricing, and execute micro-savings strategies without direct user input, requiring robust security and user-control frameworks. For B2B, the shift is towards self-governing contracts and programmatic cash management: companies will deploy agents to handle complex supply chain payments, automate escrow release upon verifiable completion of service milestones, and dynamically manage intercompany transfers, leading to significant reductions in working capital friction and requiring sophisticated interoperability standards to connect diverse corporate ledgers.
Ronen Assia
Ronen Assia is a Managing Partner at Team8 where he builds and invests in Fintech companies
AI Meets Organizational Intuition
A massive share of financial services still runs on human judgment: credit committees, fraud reviews, underwriting exceptions, compliance escalations, risk overrides. These decisions are not purely data-driven; they rely on experience, context, institutional norms, and “how we do things here.” This organizational intuition varies from one organization to the other and has been extraordinarily difficult to capture in systems, which is part of the reason why automation in financial services has plateaued at rules engines and narrow models. The result: slow scaling, inconsistent outcomes, heavy reliance on senior experts, and a persistent tradeoff between growth and risk.
AI changes this equation. For the first time, models can learn not only from structured data and written policies, but from the unwritten rules: past decisions, edge cases, reviewer behavior, internal discussions, and historical overrides. This enables a new class of AI agents that reason the way an institution reasons, applying policy through the lens of its unique risk appetite, culture, and decision-making philosophy. Rather than enforcing a one-size-fits-all model, we can build solutions which become a bank-specific decision engine, powering human-like judgment with machine-scale consistency, explainability, and speed.
The implications go far beyond operational efficiency. Capturing and operationalizing institutional judgment allows financial institutions to grow without diluting risk standards, onboard and upskill junior teams faster, reduce bias and variance in decisions, and create defensible differentiation where it actually matters. Commercial credit is a natural starting point, but this pattern will extend across fraud, compliance, and beyond. In 2026, the winners in fintech won’t just automate workflows, they’ll encode the personalized judgment itself.
Galia Beer Gabel
Galia Beer-Gabel is a Managing Partner at Team8, where she builds and invests in Fintech companies.
The Rise of Vibe Shopping
For decades, e-commerce has functioned like a catalog: a system of filters and facets optimized for direct search, not discovery. It works when consumers know exactly what they want, but breaks when they’re really searching for an outcome – a look, a feeling, a fit, a ‘vibe’ or a solution. In those moments, people still want what the internet never offered at scale: guidance, reassurance, taste and context. In 2026, that limitation breaks. Autonomous AI agents trained on billions of data points turn decision support and sales guidance into programmable, scalable software.
This marks a new era in e-commerce. Instead of static interfaces and predefined funnels, consumers engage in adaptive conversations with AI agents that interpret intent, emotion, reduce uncertainty, and guide decisions from the first question to the final purchase.
We move from attribute-based search (filtering by “red” or “size 10”) to intent-native dialogue. In practice, this introduces a new layer in the commerce stack: AI-native sales or service agents that sit between consumers and products and are hyper-personalized to the end customer, replacing static UX with conversation dialog. These agents don’t just recommend – they interpret hesitation, surface tradeoffs, build confidence, and move shoppers from uncertainty to commitment. The impact is structural: luxury and expensive categories that historically depended on human sales for trust and explanation (cars, homes, insurance, financial products) become truly online, while everyday e-commerce sees step-function gains in conversion and retention. The winners won’t be brands or marketplaces alone — they’ll be the platforms that own the decision layer, turning judgment, trust, and guidance into scalable infrastructure.
Alon Huri
Alon Huri is a Managing Partner at Team8 where he builds Fintech companies.
Banks don’t have an AI data problem, they have an AI depth problem.
An investment opportunity I’m watching closely heading into 2026: AI agents for banks – not generic agents repackaged for finance, but agents that deeply understand bank products and how bankers engage customers.
Everyone knows banks are sitting on massive amounts of data. The harder, and far more valuable, part is the soft data: call recordings, CRM notes, sales playbooks, win/loss context, and the signals that show how great bankers uncover motivation, frame value, and match the right product to the right moment. Until recently, AI simply couldn’t learn at this depth, which is why it stayed stuck in shallow use cases. That’s starting to change.
Adoption is already real. 2025 marked the shift from experimentation to deployment, but it’s still early. 77% of banks have launched or soft-launched GenAI, and ~75% of mid-sized banks ($50–250B in assets) are already live or in the pipeline. Most agents today still sit at the service layer (call centers, FAQs, ticket deflection, basic workflow optimization) not because that’s where the value ends, but because it’s the safest place to start.
The real upside comes next. As banks build confidence in controls, traceability, and agent behavior, AI moves into engagement and sales – understanding intent, framing the right product, and scaling the judgment of the best relationship managers. We’re already seeing two paths among AI agent companies: some pushing for broad scale with shallow learning, and others going much deeper, building real domain expertise in how banks interact with customers at their best. This shift won’t be led by the largest banks, who will mostly build in-house, but by the global mid-market, where coverage is thin, complexity is high, and leverage really matters. That’s where the ROI step-change happens.By 2027, the question won’t be whether banks use AI agents, but which approaches are moving beyond service into real engagement and sales at scale.
Hadar Siterman Norris
Hadar Siterman Norris is a Partner at Team8 where she invests in Fintech companies.
Digital Health
The end of operational drag in healthcare
For decades, we’ve talked about technology transforming healthcare. But something different is happening now. We’re finally feeling real pull from the market.
Pharmacists buried under prior authorization calls. Clinical trial coordinators drowning in manual data and site monitoring. Nurses managing chronic disease populations one phone call at a time. Different settings, same pattern: repetitive, operational work consuming highly trained professionals. The burden is real, and it’s keeping clinicians from doing what they trained for.
By 2026, AI-powered automation will begin to fundamentally reshape these workflows at scale. Not only by replacing clinical judgment, but by taking on the repeatable coordination that surrounds it. As patient data, engagement signals, and protocols become easier to interpret in real time, systems can handle routine actions automatically and escalate only what truly requires human expertise.
The real opportunity isn’t just time and cost savings – it’s what becomes possible when you lift the burden from the humans on the other end. Automate the repeatable, and you create space for the irreplaceable: the pharmacist who can finally counsel a confused elderly patient, the trial coordinator who can focus on complex protocol deviations, the nurse who can practice at the top of her license.
We’re not eliminating jobs. The transformations I expect in 2026 won’t just redesign how work flows through healthcare, it will put the right person in the right place at the right time.
Varda Shalev
Professor Varda Shalev is a Managing Partner at Team8 where she builds and invests in digital health companies, a member of the Board of directors of Teva and a practicing physician.
Healthcare’s missing interface
The average employee has access to more health benefits than ever: medical, dental, vision, mental health, FSAs, HSAs, wellness stipends, pharmacy benefits, telehealth. Yet most people couldn’t tell you what they actually have, let alone how to use it. Benefits enrollment happens once a year, the details vanish into a PDF, and employees are left navigating a maze of portals, phone trees, and fine print. As new models like ICHRA shift more choice (and complexity) to employees, this problem only intensifies. We’re moving toward a world where individuals must compare, select, and manage their own coverage – but without tools that let them understand, access, and pay for care in one place. In 2026, we’ll see the rise of unified benefits interfaces that finally make sense of this fragmented landscape.
The other missing interface is even more fundamental: the conversation itself. Sixty-five percent of patient engagement already happens over the phone: calls to schedule appointments, check on claims, refill prescriptions, ask about symptoms. And with AI, that volume is about to explode. As AI-powered agents handle more patient interactions via voice, chat, and email, for example, the number of conversations will grow dramatically. These interactions contain clinical signals, social determinants, operational friction. Yet almost none of it is captured. EHRs document outcomes, not the messy reality of how patients actually engage with the system. Conversations outside the physician’s office, in every medium, represent healthcare’s largest untapped data source. AI can now listen, understand, and act on these interactions at scale, turning the most human touchpoint into the most valuable one.
Assaf Mischari
Assaf Mischari is a Managing Partner at Team8, where he builds and invests in companies in the Digital Health domain.
From AI Governance to AI Operations
Right now, healthcare is building AI governance. 66% of academic medical centers have committees to approve predictive AI. Onboarding reviews. Risk assessments. Policy frameworks. This is the emerging category.
But governance assumes AI is a tool you evaluate once and monitor periodically. When AI is triaging patients, processing prior auths, and answering calls 24/7, it becomes mission-critical infrastructure. And infrastructure needs operations, not just governance.
What comes next is AI Operations Centers. Real-time monitoring of generative models and use cases in production. Uptime tracking, hallucination detection, drift alerts, response times, compliance logging. Think Security Operations Centers, but for clinical AI.
Every health system needs its own framework calibrated to patient population, payer mix, and risk profile. When AI goes down, patients don’t get triaged. That’s not a governance problem. That’s an operational imperative.
The companies building operational infrastructure for AI-native care delivery are building the next essential layer of the stack. The operating model here mirrors SOCs: not pure software, but tech-enabled services. 24/7 monitoring requires people, incident response protocols, escalation procedures. The winning model will be platforms that combine software infrastructure with service layers that can operate, respond, and adapt in real-time.
Dror Grof
Dror is a Partner at Team8, where he builds and invests in Digital Health companies.
It’s Not the AI Model, It’s the Business Model
A lot of conversations about AI start in the same place: the model. How powerful is it? How accurate? Is it generic or proprietary? While those are indeed important questions, in highly regulated high-stakes industries like fintech and healthcare, they’re rarely the ones that will determine success. What matters more, is how the technology is delivered: the business model, the workflow it fits into, and the role humans are expected to play alongside it.
For example, consider how people experience decision-making in finance versus healthcare. As consumers, do we really care whether a loan was approved by an AI model or a human underwriter? As long as the process feels fair and the outcome makes sense, it doesn’t really matter whether a human or a machine made the call. Now think about health-related decision making. Would you trust a diagnosis generated 100% by an AI model or would you still ask a human doctor to review it? Even if an AI model can analyze medical records with impressive accuracy, most patients still expect a human doctor to review their case and make the final decision.
This difference in expectations has a big impact on how AI products succeed in these types of use cases. The same underlying technology can thrive in one context and struggle in another, simply because the delivery model doesn’t match what users are ready to accept. In healthcare especially, many of the most effective AI deployments happen behind the scenes. The “frontend” experience stays familiar: care is still delivered by clinicians, services still look like services. What changes is how those professionals work. From the patient’s point of view, nothing radical has happened – and that’s often exactly the point.
For companies building AI in fintech and healthcare in 2026, this can be a useful lens. Strong technology matters, but it’s rarely sufficient on its own. Adoption depends on understanding where trust lives, how decisions are perceived, and how innovation can be introduced without breaking the implicit contract between provider and user.
Asaf Charnilas
Asaf is a Principal at Team8, having previously held multiple managerial roles at hyper-growth startup companies.
GTM
The Next Generation of Category-Defining Companies Will Start as Infrastructure
Across AI, cybersecurity, fintech, and digital health, the most important companies of the next decade won’t start by selling a full solution. They’ll start with a narrow but critical infrastructure wedge: something deeply technical, often unglamorous, that quietly becomes indispensable. These companies won’t lead with dashboards or workflows. They’ll embed themselves into the system of record, the system of control, or the system of trust, becoming impossible to remove once adopted.
We’re already seeing the pattern repeat. In AI, it’s evals, observability, and governance layers. In security, it’s reliability, coverage validation, and identity for non-human actors. In fintech, it’s agent-native payments, risk, and reconciliation infrastructure. In healthcare, it’s the infrastructure that makes care delivery economically viable rather than administratively heavy. Once these infrastructure layers become foundational, they naturally expand upward into platforms, workflows, and ultimately full-category ownership. The winners won’t look big on day one, but by 2026 and beyond, they’ll define entirely new categories precisely because they started where the real leverage lives.
Sarit Firon
Sarit Firon is a Managing Partner at Team8, where she invests in Digital Health companies.
Authenticity is the new moat
AI can now generate unlimited content, blogs, ads, social posts, even videos, at near-zero cost. Every company can flood every channel.
The result? Noise is infinite. Attention is scarce. And audiences are increasingly skeptical of anything that feels synthetic.
In 2026, brand becomes the antidote. Not brand as a logo or a tagline, but brand as trust, earned through consistency, authenticity, and real human connection. The companies that win will be the ones that feel real in a sea of generated content.
That means founder voices, genuine community, transparent communication, and a point of view that can’t be templated, automated, or outsourced to a prompt.
Paradoxically, the explosion of AI-generated content makes early investment in authentic brand-building more valuable than ever. When anyone can say anything, founders who treat brand as a pre-seed priority, not a Series A or B afterthought, will command premium positioning, attract better talent, and close deals faster.
All of it built on one thing that AI still can’t generate at scale: trust.
Asaf Azulay
Asaf Azulay is a Partner and the CMO of Team8.
