The distinction at the heart of this year’s theme is one that separates two generations of artificial intelligence tools. The generative AI systems that dominated the previous funding cycles respond to a single prompt: a researcher asks a question, the model produces an answer, and the interaction ends. Agentic workflows operate on an entirely different architectural principle. An agent is given an objective and then plans a sequence of steps, executes those steps by calling external tools and services, inspects the results, revises its plan when necessary, and loops through this cycle until the work is complete. In a research context, that might mean an agent that retrieves datasets from a repository, cleans and reconciles them against institutional records, drafts an analysis, checks its own output against the source material, and then prepares a submission-ready package. Each of those stages involves tool calls, intermediate state, and decisions that were previously made by a human sitting at a keyboard.

Steve Scott, VP Portfolio Development at Digital Science, framed the shift in terms that will resonate with anyone who has watched the AI tooling landscape evolve. “AI tools are increasingly able to act on research, not just describe it,” he said at the launch. “The next breakthroughs will come from agentic workflows, autonomous and multi-step, that researchers, institutions and funders can actually rely on.” His second observation is arguably the more consequential one for the sector. “This is a different category from the generative AI tools we were funding three years ago,” Scott explained. “An agentic workflow plans, executes and reviews multi-step work rather than answering a single prompt.” The distinction matters because autonomy changes the risk profile completely. A chatbot that hallucinates a citation is an inconvenience; an autonomous agent that hallucinates a citation and then acts on it, submitting a flawed manuscript or misclassifying a grant application, embeds that error into the permanent record of research.

This is why the 2026 theme insists on the word “trust,” and why Scott was careful to define what trust means in this context. “Trust in this case means quality rather than detection,” he said. “An agent that shows its working, not a tool that catches bad actors after the fact.” The technical implications of that statement are substantial. Provenance, in the sense intended here, means that every output an agent produces carries a verifiable record of the inputs, data sources, model versions and prompts that produced it. Audit means that the full chain of an agent’s actions, every tool invocation, every intermediate artefact, every decision point, is logged in a form that a human reviewer or institutional compliance system can reconstruct after the fact. Governance means that agents operate within defined permission boundaries, escalate to human judgement at meaningful checkpoints, and remain accountable to identifiable owners within an organisation. A trustworthy agentic workflow is, in engineering terms, a system designed for inspectability rather than opacity.

The focus areas for this year’s grant map those engineering requirements onto the research lifecycle. Trusted authorship and writing refers to agents that assist with drafting, summarising and editing while maintaining an unambiguous account of what was contributed by a human, what was contributed by a machine, and on what basis. Enterprise research workflows covers agents operating at institutional scale, interfacing with the systems that universities and research organisations actually run, from current research information systems to repositories and grant management platforms. Trusted decision agents, perhaps the most delicate of the four categories, addresses systems that support funding and institutional decisions, where the stakes of an unexplainable or unverifiable output are at their highest. Trusted publication workflows rounds out the list, encompassing the journey from manuscript preparation through peer review support to publication, with provenance intact at every stage. Digital Science notes that novel applications of agentic AI benefiting any part of the research lifecycle will be considered, with these four areas serving as indicative rather than exhaustive territory.

The eligibility criteria are deliberately forgiving, which has always been part of the programme’s appeal to early-stage builders. Applicants do not need revenue, a finished build or a complete business plan. A prototype, a working product or a well-formed concept is enough to enter. What the application must articulate is the problem being solved, the approach taken, why the solution fits an agentic and trustworthy workflow, and where the work stands today. That last requirement is telling. In a funding landscape crowded with AI pitches, Digital Science is effectively screening for teams that understand the difference between a demo and a system an institution would actually deploy. The equity-free nature of the award is equally significant for early-stage teams: £25,000 arrives without the recipient surrendering ownership, allowing founders and academic groups to use the capital as validation capital rather than as the beginning of a dilution cycle.

The timing of the theme reflects a broader inflection point in the research technology sector. Tool-calling interfaces and orchestration frameworks have matured to the point where multi-step autonomous systems are technically feasible for small teams, not just large laboratories, and the last two years have seen an explosion of experimental agents across science, from literature-screening assistants to automated data-analysis pipelines. What has lagged behind is the institutional infrastructure of trust. Universities, funders and publishers operate under obligations of research integrity, data protection and auditability that generic consumer AI tools were never designed to satisfy. An agent that cannot explain where a piece of information came from, or that leaves no traceable record of its actions, is functionally unusable inside those environments regardless of how capable its underlying model is. The 2026 Catalyst Grant is, in effect, a bet that the teams who solve the trust problem, rather than the teams who chase raw capability, will define the next generation of research infrastructure.

Scott also pointed to the value the programme delivers beyond the cheque itself, a point that previous cohorts have repeatedly confirmed. “Catalyst Grant’s real value lies beyond the funding itself,” he said. “It’s in the opportunities provided to winning teams: the conversations it starts, practical advice, introductions to other experts in the field, and a sharper idea of what successful innovation looks like.” Over sixteen years, the programme has functioned as an early signal of where research tooling is heading, and its alumni network has become a genuine asset for teams navigating the difficult passage from prototype to product. For researchers and founders working on agentic systems, access to a company whose portfolio spans research information management, altmetrics, data repositories, authoring platforms and patent services offers a vantage point across the entire ecosystem that few accelerators can match.

For those considering an application, the practical details are straightforward. Full eligibility criteria and application instructions are available on the Catalyst Grant website, and questions about the programme can be directed to catalyst@digital-science.com. The award is open globally to individuals, startups and research teams with early-stage ideas, and Digital Science is encouraging applicants and observers to join the conversation on social media under the hashtag #CatalystGrant. The company itself sits at the centre of the ecosystem this year’s theme addresses: its brands include Altmetric, Dimensions, Figshare, IFI CLAIMS Patent Services, metaphacts, Overleaf, ReadCube, Symplectic and Writefull, collectively covering research evaluation, data publication, authoring and institutional workflow management. That breadth gives the 2026 theme particular weight, because the company is not merely observing the shift toward agentic AI; it operates the infrastructure such agents would need to work within. Media enquiries can be directed to David Ellis, Manager of Media and PR at Digital Science. As autonomous systems move from novelty to necessity in research, the question the 2026 Catalyst Grant poses is no longer whether agents will act on science, but whether the people building them can make their actions visible, verifiable and worthy of institutional confidence.

News Publication Date: 1-Sep-2026

Web References: Digital Science Catalyst Grant

References: Digital Science’s 2026 Catalyst Grant seeks agentic AI workflows you can trust. Available at: https://www.digital-science.com/about-us/investment/catalyst-grant/

The framework is based on a simple but powerful idea: complex intelligence may be easier to create when it is assembled from many specialized forms of limited intelligence. Rather than asking one enormous AI model to understand and solve an entire problem at once, COIN divides the overall objective into subtasks that can be analyzed, executed and verified separately. The results are then recombined into a coherent solution that satisfies the requirements of the original task.

Yang Yuan, an associate professor at the Institute for Interdisciplinary Information Sciences at Tsinghua University and corresponding author of the study, compared the concept with classical calculus. Calculus can determine the area beneath a complicated curve by dividing it into many smaller sections and adding their contributions together. In a similar way, COIN seeks to represent a complex intelligent process as a structured composition of simpler operations.

The mathematical foundation of the framework is a structure known as a Grothendieck topos. In mathematics, a topos can be understood as a formal environment that describes objects, relationships and the rules governing how they interact. For artificial intelligence, the researchers use this setting to specify what each component of a system is allowed to observe, which conditions it must satisfy and how it should exchange information with other components.

Yuan likened the arrangement to the design of a large aircraft. Engineers responsible for wings, engines, navigation and control systems work on different parts of the airplane and do not need access to every detail of the entire project. Each group operates within a limited local view, but the components must still fit together at shared boundaries. The wing must connect correctly to the fuselage, the engines must interact with the control system and all parts must meet common safety requirements.

In COIN, these local views correspond to specialized subtasks or agents. The framework establishes the interfaces through which they communicate and defines how their solutions can be combined without violating global constraints. This is intended to address one of the central challenges facing agentic AI: coordinating independent systems while preserving consistency, reliability and accountability across the complete workflow.

The researchers describe the process using the language of decomposition and recomposition. Decomposition breaks a broad objective into smaller problems that are sufficiently well-defined for individual models or agents to solve. Recomposition then assembles those local results into a global outcome. The mathematical rules are important because simply joining the outputs of multiple AI systems can produce contradictions, duplicated work or failures at the points where their responsibilities overlap.

According to Yuan, the framework reflects a broader view of intelligence in which intelligence is inseparable from structure. A system may appear incomprehensible because its organization is hidden, rather than because its individual parts are intrinsically impossible to understand. If the correct structure can be identified, a seemingly overwhelming problem may become a collection of bounded tasks that can be tested independently. This could make future AI systems easier to inspect and verify, particularly when they are deployed in high-stakes environments.

The proposal also challenges the idea that progress in AI must depend on creating a single model with unlimited capabilities. Many smaller systems, each possessing a narrow but useful form of intelligence, could potentially be organized into networks capable of solving problems beyond the capacity of any one model or human expert. Such systems might distribute planning, perception, reasoning, coding and monitoring across specialized agents while using formal interfaces to maintain cooperation.

COIN is presented as an early step toward a common mathematical language for intelligence. The researchers aim for such a language to describe what AI models learn, how complex tasks can be divided and how numerous limited intelligences can be coordinated into larger systems. The work, co-authored by Andrew Chi-Chih Yao, professor and dean of Tsinghua’s Institute for Interdisciplinary Information Sciences, appears in the journal iFuture under the title “Calculus of intelligence: A topos-monadic framework for agentic workflows.” If the framework can be translated into practical engineering tools, it could influence how future AI systems are designed—not as solitary superintelligences, but as carefully coordinated communities of specialized agents.

Subject of Research:
A mathematical framework for decomposing and recomposing complex artificial intelligence tasks, particularly in agentic AI workflows.

Article Title:
Calculus of intelligence: A topos-monadic framework for agentic workflows

News Publication Date:
17-Jul-2026

Web References:
https://doi.org/10.26599/IF.2026.9710001
https://www.sciopen.com/journal/3135-3169

References:
Yang Yuan and Andrew Chi-Chih Yao, “Calculus of intelligence: A topos-monadic framework for agentic workflows,” iFuture, DOI: 10.26599/IF.2026.9710001.

Image Credits:
iFuture, Tsinghua University Press

Keywords

Artificial intelligence, agentic AI, calculus of intelligence, COIN, Grothendieck topos, mathematical AI, AI agents, multi-agent systems, task decomposition, intelligent systems, Tsinghua University, AI workflows