We’re only a few weeks into 2026 and the conversation around AI is already shifting. The focus is no longer on eye-catching demos or speculative promises, but on systems that deliver real results in production — at scale and with durable economics. To take a pulse of what’s coming next, we spoke with investors and operational experts across the Battery team about what changes they expect to define the year ahead.

Their predictions point to a more pragmatic phase of the AI cycle. Agentic systems will mature, but not always in the ways hype has suggested. Persistent memory, system-level reasoning and continuous learning in production will reshape how software is built and used. At the same time, companies will grapple with cost, data ownership and performance, driving renewed interest in open-source models, self-hosting and infrastructure as a source of durable differentiation. Below is a snapshot of what we expect 2026 to bring.

I. AI in production

“Early days of agentic applications have been dominated by search and coding use cases. 2026 will see the expansion of agentic AI in enterprises and proliferation of go-to-market use cases aimed at effective prospecting and efficient sales motions.” — Dharmesh Thakker, General Partner (LinkedIn)

“Inefficiencies in demand gen, prospecting and sales execution have increased S&M as a percentage of revenue in many industry sectors. Agentic AI systems like Clay, Unify*, Gong*, Apollo*, 1Mind* and others are achieving high efficiency and accuracy in targeting specific buyers with personalized messaging and streamlined selling. We expect to see tech forward enterprises adopt these approaches at scale in 2026, and the formation of a new GTM stack that could upend SaaS behemoths like Salesforce and Marketo.”

“The new, primary language for writing software will be English-language prompts to LLMs.”— Max Schireson, Operating Partner (LinkedIn)

“This is good news and bad news: much easier to write code… but how the heck do we know if it works correctly? We slowly begin to rethink how product definition, coding and testing come together to create not just more software, but useful and correct software.”

“By the end of 2026, AI coding tools must graduate from coding features to building software that takes into account the greater software, systems & infrastructure – or risk hitting a ceiling in how useful they can be.” — Scott Goering, Business Development Partner (LinkedIn)

“System-level comprehension and reasoning will be critical: tools must work with/on core product codebases (many of which are decades-old legacy systems) and factor in system constraints and real-world operating conditions before executing resource-intensive operations.”

“The agentic-only heat will cool and UI-based SaaS will come back into favor.” — Mikey Hoeksema, Principal (LinkedIn)

“As enterprises hit reliability, governance and change-management limits, agents sold as pure automations will lose appeal relative to AI embedded in familiar workflows. Durable companies will look more like SaaS companies of old with AI superpowers than collections of loosely orchestrated agents.”

“If coding assistants were the breakout enterprise-AI use case of the last cycle, 2026 will belong to the back office.” — Sudhee Chilappagari, Principal (LinkedIn)

“Across HR, finance, IT, procurement, legal and supply chain, teams are chronically understaffed and buried under repetitive, rules-heavy workflows. This is fertile ground for agentic AI.

We expect AI agents to take over workflows like accounts payable and receivable, cash application, IT support-ticket resolution, vendor onboarding, contract review and procurement negotiations. These tasks are structured enough to automate, yet complex enough to benefit from reasoning models that can interpret context, handle exceptions and learn over time. Just as importantly, the ROI is unusually clear. This year, AI will quietly—and decisively—run the operational backbone of the modern enterprise.” 

II. Market dynamics and economic shifts

“IPO markets have opened, but the 2025 cohort has been largely underperforming. 2026 marks the launch of $100B+ enterprise value companies going public and providing institutional investors an outlet to participate in a new cohort beyond the overplayed Magnificent Seven.” — Dharmesh Thakker, General Partner (LinkedIn)

“In 2026, data moats will become real as application companies increasingly post-train open-weight models on proprietary data.” — Jason Mendel, Vice President (LinkedIn)

“Infrastructure decisions will drive product differentiation, with fine-tuning and reinforcement learning enabling companies to customize off-the-shelf models to their specific domains and workflows. Rather than batch training jobs, agents will continuously learn from new data, creating closed-loop systems that evolve in production and improve over time.”

“Diminishing returns to training scale will catalyze innovation in different model architectures, opening up space for new entrants to compete on cost and specialized reasoning vs. the current super-capitalized leaders.” — Marcus Ryu, General Partner ( LinkedIn)

“If 2024 was about experimenting with AI and 2025 was about shipping it, 2026 will be about making it pencil out.”— Sudhee Chilappagari, Principal (LinkedIn)

“As AI moves from novelty to necessity, we expect a decisive shift toward open-source and self-hosted models as companies look to regain control over three things that now define AI ROI: data, cost and performance.

Self-hosting is no longer just about privacy or compliance. It’s about tuning models to specific workflows, reducing inference costs, and avoiding the tax of per-call pricing at scale. As tooling around deployment, orchestration and observability matures, running your own models will move from “only for the infra-savvy few” to a mainstream architectural choice. The winners in 2026 won’t be the teams with the flashiest demos. They’ll be the ones who can scale agentic systems sustainably—without hemorrhaging margin or handing over their most valuable data.”

“The AI industry finally realizes that throwing money at the problem by training bigger models with more data isn’t how AI will ‘grow up’.” — Max Schireson, Operating Partner (LinkedIn)

III. Emerging technologies and new frontiers

“Voice and video AI will continue to astound us and transform many horizontal domains such as physical security, contact centers, retail and industrial operations, and data collection. Agentic AI will continue to lag behind its promise to automate all digital work but will become steadily more useful with better integrations to enterprise systems.” — Marcus Ryu, General Partner (LinkedIn)

“Open-source and open-weights models like Reflection*, Llama and Mistral from the U.S. and Europe will provide viable alternatives to Chinese models in 2026.” — Dharmesh Thakker, General Partner (LinkedIn)

“Combined with data platform vendors like Databricks* providing agentic building blocks and reinforcement learning, enterprises will build fine-tuned agentic applications that drive their differentiated edge.”

“I see 2026 as the year AI shifts from maximizing benchmark performance to becoming a true teammate in the workplace.”— Brandon Gleklen, Principal (LinkedIn)

“A big part of this shift will be driven by advancements in memory, but two less-discussed factors will matter just as much: AI integration into collaboration systems (such as joining a Zoom call, speaking only when appropriate, screen-sharing content) and improvements in AI personality that make AI colleagues genuinely enjoyable to work with. “Personality Engineer” will emerge as a job title in AI companies.”

“In 2026, AI systems shift from today’s stateless, prompt-driven models to architectures that internalize experience through persistent state, online adaptation and intent inference.” — Lior Mallul, Principal (LinkedIn)

“As orchestration moves out of the user workflow and into the model and product layer, software is rebuilt around continuous learning rather than discrete interactions. The result is a platform reset across enterprise and consumer markets, with value accruing to systems that learn in production.”

“Defense adoption of AI-enabled continuous test and validation tooling will shift from fringe to mission-critical in 2026 as the DoW demands faster development and iteration cycles from our defense industrial base — unlocking a new class of repeatable, certifiable, data-first platforms as investable defense infrastructure.” — Aaron Neil, Vice President  (LinkedIn)