When many agents could do a task, the system usually picks one silently at build time, collapsing the real trade-off between cost, speed, privacy, and quality into a default nobody chose for this specific job.
← Back to patterns
Intelligent AI Delegation Agent2Agent (A2A) Protocol
Agent Workflow
Research PatternTask Bid Market
Let the Person Pick the Bid, Not the System
When many specialist agents could handle a task, most systems silently route to one chosen at build time — before anyone knew this job would be high-stakes. The move: broadcast the task, collect competing bids, and lay them out side by side — cost, time, privacy, reputation — with trade-off controls, so the person picks on the actual trade-off. For an open agent web, this is where routing is heading, not where it ships today.
Metaphor
Open Tender
Interaction
Broadcast RFQ, Receive Bids, Pick by Trade-Off
Use case
Routing to specialist agents, Per-task provider choice, Cost-vs-privacy trade-offs
Failure mode
Design-Time Dispatch
Control mechanisms
Task RFQ Broadcast · Bid Objects (cost/duration/privacy) · Reputation Filter · Smart Contract Settlement
Reference examples
A2A protocol · AP2 Mandates · EigenLayer AVS operator marketplace · Upwork open bidding (human analogue)
Concept Frame
Hierarchical dispatch assumes you know the worker. The agentic web does not.
Hierarchies work when you can pick the right agent up front. An open web of specialist agents breaks that: hundreds of trade-offs — cost vs latency vs privacy vs reputation — and no single right answer per task. So make assignment a comparison surface: the task goes out, agents bid, and you read the bids side by side and pick on the trade-off that matters for this job. The protocol underneath (signed bids, a stake against fake agents, settlement) is real work, but the UX move is simple — you stop picking an agent by name and start reading bids. The open agent web this assumes is still emerging.
Interactive Example
The live side-by-side is the comparison.
This prototype already contains the contrast between the status-quo approach and the pattern response. The left column names the failure mode of the naive baseline (fixed dispatch, label-as-done, hidden hierarchy, uniform friction); the right column is the move.
Pick the delegatee you know, or let the market surface the one you don't.
Hierarchical delegation assumes the delegator already knows who to hand the task to. Across an open agentic web of thousands of specialist agents, that assumption collapses, and ranking candidates by reputation alone flattens a multi-axis trade-off into a single number — cost, latency, and privacy all crushed into one score that hides what you actually gave up.
Task Bid Market makes assignment a comparison surface: the task goes out, competing agents bid with their cost, time, privacy, and reputation, and you see them side by side instead of sorted into one ranking. You pick the bid that fits this job, and the choice carries its reasoning with it — which axes it won on, what you traded away. The hard parts that make it safe — a stake against fake agents, settlement — sit under the surface.
No one can keep an accurate model of every agent out there — so let the candidates price themselves, and compare the offers instead of guessing.
DOPE Lens
How this pattern improves AI interaction quality.
Observability
Every bid is visible side-by-side — cost, duration, privacy guarantee, reputation portfolio, and the agent's prior task history
Directability
Delegators set the trade-off priorities (latency vs cost vs privacy) and the market filters; pick is one click on the chosen bid
Predictability
Bid contracts specify scope, schedule, and penalty for default before commitment; reputation bonds make defection costly
Explainability
The selected bid carries its own justification — which axes it dominated, which competitors it Pareto-replaced
Fit
Where it earns its place.
Use it
when
when
Surfacing the candidate bids and letting the person decide on the trade-off beats a silent build-time default, because no one choosing months ahead could know which axis — cost, speed, privacy, or quality — this particular task would turn on.
Avoid
when
when
The right agent is known and stable, or the task is routine and cheap — then a fixed route beats running an auction for every job.
Closest neighbours
No product ships this exactly — these come closest.
This is a research pattern — we describe a UX move that no shipping product implements cleanly today. The entries below are adjacent: they share the spirit of the pattern even if they take a different shape.
A2A protocol
Linux Foundation (orig. Google)
Peer-to-peer agent transport layer with Agent Cards (capability and skill manifests) and task objects (lifecycle states). The infrastructure foundation for marketplace-style delegation between agents on the open web.
Visit Linux Foundation (orig. Google) ↗
AP2 · Mandates
Google · Google Pay
Cryptographically signed intents authorising an agent to spend funds or incur costs on behalf of a principal. Provides the financial liability layer the bid market needs — a delegator can issue a mandate ceiling, bidders post stake against it.
Visit Google · Google Pay ↗
EigenLayer · AVS operator marketplace
Eigen Labs
Game-theoretic open market from the web3 stack: operators stake to take on verification work and get slashed for faulty execution; delegators choose operators on stake, performance, and track record. The closest live staking-and-selection market after TrueBit's protocol was exploited and pivoted away (Jan 2026).
Visit Eigen Labs ↗
Upwork · open bidding (human analogue)
Upwork
The mature human analogue. Clients post tasks with budget ranges; freelancers submit proposals with rate, timeline, portfolio link, reputation score. Hiring decision is a Pareto trade-off across cost/quality/speed — the same shape as the agentic-web pattern, applied to humans for two decades.
Visit Upwork ↗
Research Frame
Why this pattern exists, and where it breaks.
Treat the assignment as an open set, not a fixed roster: broadcast the task, collect competing bids, lay them out side by side — cost, duration, privacy, reputation — and let the person pick the bid that fits this task and see what each pick costs them.
Chat commits to a single agent up front, in the system prompt and tool wiring, and has no surface for comparing an open set of candidates and choosing per task at the moment of need.
The hard part sits under the comparison UI: without stake-on-bid and Sybil resistance an attacker floods the board with fake agents, without scoped requests sensitive task details leak to losing bidders, and without portable reputation the market quietly favours incumbents.
Common questions
What people ask about this pattern.
When should I use the Task Bid Market pattern?
+
Surfacing the candidate bids and letting the person decide on the trade-off beats a silent build-time default, because no one choosing months ahead could know which axis — cost, speed, privacy, or quality — this particular task would turn on.
When should I avoid the Task Bid Market pattern?
+
The right agent is known and stable, or the task is routine and cheap — then a fixed route beats running an auction for every job.
What problem does Task Bid Market solve?
+
When many agents could do a task, the system usually picks one silently at build time, collapsing the real trade-off between cost, speed, privacy, and quality into a default nobody chose for this specific job.
Why is chat the wrong fit for this?
+
Chat commits to a single agent up front, in the system prompt and tool wiring, and has no surface for comparing an open set of candidates and choosing per task at the moment of need.
Grounded in Research
Not just opinion.
The papers, guidelines, and usability studies this pattern rests on.
Nenad Tomašev, Matija Franklin, Simon Osindero
Google DeepMind · 2026
Framework for safe and scalable delegation: cognitive friction, authority gradient, verifiable completion, and task-bid markets as the missing UX layer for cross-org agent work.
Google, now Agentic AI Foundation (Linux Foundation)
Industry standard · 2025
Open spec for inter-agent communication with optionally signed Agent Cards (JWS). It signs who an agent is, not what it delivered; task-outcome attestations are still open community proposals.
Related Patterns
Curated connections, not just neighbors.
Hand-picked relationships — what this pattern pairs with in practice, what it replaces, what it builds on, and what builds on it.
Often paired with
Verifiable Task Completion
A winning bid promises an outcome; Verifiable Task Completion proves it was delivered before you accept the work.
Alternative to
Supervisor-Worker
In Supervisor-Worker the supervisor assigns each piece to a specialist; in Task Bid Market you pick among competing bids.
Alternative to
Persona Switcher
Both let you pick who takes the task: Persona Switcher from a set of named lenses, Task Bid Market from competing bids.