A position paper for a 2026 conference workshop on how people and AI agents work together. It reviews research on explicit goals and commercial goal tools, with meetings as a worked example.
Goal researchDraws on goal-setting theory and Objectives and Key Results, a practice of setting objectives with measurable results.
Goal toolsExamines commercial task managers, project platforms and goal-setting products, such as Asana, ClickUp and Atlassian Goals.
Meetings exampleUses earlier research on meeting goals to show what stating goals upfront can achieve.
RequirementsArgues towards six design requirements for systems in which people and AI agents work from shared goals.
BWhat the paper argues
Why goals matter more with AIthe argument from goal research
•The paper argues explicit goals focus attention, aid coordination, sustain effort and benchmark results.The paper draws on goal-setting theory and Objectives and Key Results practice. It argues AI amplifies each role, because cheaper production raises the cost to human attention.
•The paper argues earlier goal systems failed because they demanded too much precision.Group support and workflow systems needed exact goals, while real goals are fuzzy and change. The paper argues generative AI can work with goals stated loosely.
What current goal tools missthe paper's review of commercial goal tools
•In the paper's review, no goal tool helped people work out what they wanted to achieve.A few tools offer goal objects, alignment hierarchies and status rollups. The tools' AI help only rewords goals.
•In the paper's review, no goal tool used goals across apps or to direct AI agents.Goals stayed inside single products and out of sight while people worked.
What to builda worked example and design requirements
•Meeting research cited by the paper found stating a purpose upfront led to better agendas.Prompting people to state purpose and success when creating a meeting also led to more focused meetings. The paper argues AI agents could work from such goals.
•The paper proposes six requirements for systems where people and AI agents share goals.Help to articulate goals, persistent shared goals, signs of alignment, goal history, simulated decisions and goal cues in daily work. AI agents would report progress against goals.
CKey takeaway
Treating goals as first-class abstractions, shared and explicit, can help people and AI agents aim at outcomes rather than outputs. Generative AI may make this practical by working with fuzzy, changing goals.
Made for this site with AI assistance from the verified research brief. It is not the paper's own figure.
Abstract
As AI assumes more of the material production in knowledge work, human effort shifts toward planning, orchestration, and evaluation, all of which revolves around goals. Yet goals remain poorly represented in knowledge work tools and workflows: implicit, unexpressed, or confused with outputs. Beyond their importance for human work, clear goals are fundamental to human-AI communication and collaboration. We review research establishing the value of explicit goals, show through a review of commercial tools that existing ecosystems support goal tracking but not goal articulation, alignment, or contextual use, and use meetings as a proving ground demonstrating that upstream goal articulation produces disproportionate downstream value for both humans and AI agents. We argue that goals should be encoded as first-class abstractions that drive human-AI collaborative workflows and that generative AI's natural-language capabilities make this a uniquely opportune moment. We outline six design requirements for goal-oriented human-AI collaborative systems.
Research brief
By Lev Tankelevitch and Sean Rintel.
Takeaway: Goals should be first-class abstractions that people and AI agents share and work from, because as AI takes over production, human work shifts to planning, orchestrating and evaluating around goals. Generative AI makes this feasible by handling fuzzy, evolving goals, and meeting intentionality shows the payoff of articulating them upfront.
Why this matters
What was known
Goals organise knowledge work, yet they rarely appear in its tools and artefacts, and the goal tools that exist support tracking rather than articulation. Goals often exist entirely in people's minds, either unformed or unexpressed, often leading to a confusion of outputs for outcomes(§1, pp. 1–2)
What this paper adds
We bring together goal-setting research, a review of commercial goal tools and meetings as a worked example. We argue that generative AI's ability to work with natural language removes the demand for premature precision that sank earlier goal-structuring systems. Goals can now be accommodated computationally. The open question is how to treat goal representations as living resources, open to interpretation and revision, not fixed contracts.(§5, p. 4)
Who it matters to
Designers of knowledge work tools, collaborative systems and AI agent orchestration, and the many workers whose roles shift towards planning, orchestration and evaluation as AI takes on more production.
Research question
We argue that human-AI collaborative systems should support the articulation of goals (desired outcomes) and encode goals as explicit abstractions. These abstractions should drive goal-oriented experiences that shape people's and AI agents' thinking, communication and workflows towards better outcomes rather than more outputs.
Method and data
Design
Position paper for a CHI 2026 workshop. It reviews research on the value of explicit goals (goal-setting theory and OKR practice) and commercial goal tools (task managers, project platforms and dedicated goal-setting products). Meetings serve as a worked example.
Setting
Knowledge work in which human-AI teams collaborate.
Participants / data
Commercial goal tools including Asana, ClickUp and Atlassian Goals. Prior research on meeting goals.
Analysis
Argument and synthesis, leading to six design requirements for goal-oriented human-AI collaborative systems.
Findings
Explicit goals do four jobs. Research on goal-setting theory and on OKR (Objectives and Key Results) practice shows that explicit goals direct attention, let collaborators coordinate without tracking every action, drive persistence and give a benchmark for judging results. We argue AI amplifies each of these, starting with attention: #
“As AI cheapens the production of knowledge work, the cost to human attention increases”
Goal tools track, but don't help you think. In our review of commercial goal tools (task managers, project platforms and dedicated goal-setting products), a few offer goal objects, alignment hierarchies and status rollups. But none helps people work out what they are actually trying to achieve, and their AI help only rewords goals. #
“Existing tools assume goals are set once and occasionally reviewed. They are not built for a world where goals are living abstractions that shape how people and agents coordinate.”
Goals are missing where work happens. Goals in these tools stay inside single products and out of sight while people work. None carries goals across applications, shows them during the work itself, or uses them to direct AI agents. #
(§3, p. 3)
A little goal-setting pays off later. In meetings (our worked example), earlier research found that prompting people to state a meeting's purpose and what success looks like when creating it led to better agendas and more focused meetings. We argue the same encoded goals could guide AI agents, for example: #
“a recap agent can assess what was achieved rather than merely what was said, a scheduling agent can reason about whether the right people are invited”
Earlier goal systems demanded too much precision. Group Support Systems, workflow platforms and similar coordination tools tried to systematise the goals behind collaborative work and mostly failed. We argue generative AI can succeed where they did not, because it can work with goals stated loosely in natural language. #
“Most foundered on the same problem: real goals are fuzzy, fluid, and contextdependent, while the systems designed to capture them demanded premature precision”
Six requirements, including agents that report on goals. We propose six requirements for goal-oriented systems. They are help with articulating goals, goals stored as persistent shared objects, surfacing alignment and misalignment, goal history, simulating how decisions cascade, and goal cues inside everyday workflows. For AI agents this means: #
“AI agents should receive delegated goals and report progress against them, not merely against tasks completed.”
Questions this paper answers
Why do goals matter more when working with AI?
Because as AI takes on more of the production, people's work shifts to deciding what to do, who should do it and whether it is on track. These all revolve around goals. Producing more becomes cheap, so attention to the right work becomes the bottleneck. (See §2, p. 2.)
What is the difference between a goal and an output?
A goal is a desired outcome. An output is something produced, such as a meeting, a deck or a shipped feature. Because goals often stay unexpressed, people confuse the two and treat outputs as ends rather than means. Tools should help people and AI agents tell them apart. (See §5, pp. 4–5.)
Isn't writing down goals just more admin?
We argue a small upstream effort pays off downstream. Research on meeting intentionality found that prompting people to state a meeting's purpose and success conditions when creating it led to better agendas and more focused meetings. Those goals could also let a recap agent judge what was achieved. (See §4, p. 3.)
Why did earlier goal software fail, and what is different now?
Earlier systems demanded that people specify fuzzy, changing goals precisely up front, so users worked around or abandoned them. Large language models can work with goals stated loosely in natural language, ask clarifying questions and follow how goals change, so goals can stay open to revision. (See §5, p. 4.)
Contributions and implications
Design goal articulation interfaces that support reflective, iterative thinking beyond form-filling and are intelligible to both humans and AI. First, designing goal articulation interfaces that support reflective, iterative thinking beyond form-filling, and ensure mutual intelligibility for humans and AI.(§6, p. 5)
Build interoperability layers so goal representations can travel across tools and contexts.
Develop goal-oriented experiences and evaluation frameworks that steer collaboration towards outcomes over outputs.
›
Concepts
Goals
Our deliberately broad definition: We define goals broadly as desired outcomes at any altitude, from task-level to strategic, hierarchically related, and often dynamically evolving(§1, p. 1)
Goal articulation
The step that no goal tool we reviewed supports: the reflective process of working out what you are actually trying to achieve(§3, p. 3)
Goals as first-class representations
Goals stored as structured, persistent objects linked to people, agents, artefacts and workflows. Why that matters: Making goals representationally first-class is what allows them to travel across tools and contexts.(§5, pp. 4–5)
Outcomes versus outputs
Desired results versus the artefacts produced along the way. The confusion we warn against: Meetings, decks, and shipped features are treated as ends rather than means.(§1, p. 2)
How to cite
Lev Tankelevitch and Sean Rintel. 2026. Goals as First-Class Abstractions in Human-AI Collaboration. AutomationXP26 Workshop, CHI 2026.
This brief was prepared with AI assistance and published by Sean Rintel. It is not the authors' published text. Page numbers refer to this PDF.
BibTeX
@inproceedings{tankelevitch2026goals,
author = {Tankelevitch, Lev and Rintel, Sean},
title = {Goals as First-Class Abstractions in Human-AI Collaboration},
booktitle = {AutomationXP26 Workshop, CHI 2026},
year = {2026},
month = {April},
publisher = {ACM},
url = {https://matthiasbaldauf.com/automationxp26/papers/AutomationXP26_paper_0982.pdf},
}