Rohit Krishnan

Building
worlds

Multi-agent simulations

I simulate workplaces and economies to study how agents work, trade and organise themselves. This is experimental economics for the agentic economy: testing the rules and institutions we will need.

I’ve wanted to build little worlds since I was a kid. Companies, towns, societies. Now I can! And put AI agents inside them, give them things to do, and see how those worlds change.

A workbench holding miniature agents, a company, a town and a market.

Findings

Selected studies

STRATEGIES, OBJECTIVES & GOVERNANCE

Objectives and theft in a simulated town

Five models wrote strategies for households in a 16-house town. The simulator played out those strategies day by day. In a separate test, counting every occupied home in the score stopped theft.

Town study, comparisons and sources ↗
Two town runs compare family framing with a team score, and game framing with a score that counts all occupied homes.
The two runs shown here differ in both how the town is described and what the score rewards. Open full-size illustration ↗

REVIEW & PERSUASION

Persuasion in AI audits

An actor could persuade an AI auditor to mark an unsupported decision compliant without providing new admissible evidence. An independence reminder reduced that kind of false approval from 12% to 4%.

12%Ordinary audit
4%Integrity reminder

False approvals after persuasion, in a staged study of 150 cases.

Audit study and sources ↗

STRUCTURE & SCALE

Shared task boards and parallel workers

A shared task board improved inbox scores for one agent and for several working in parallel. With the board, the two groups scored the same.

Quality score after a jump in workload
Without boardWith board
One worker0.500.63
Parallel workers0.460.63

The smaller model still struggled with the board.

See the full four-way comparison ↗

ORGANISATIONS & SHARED TRUTH

Record drift in a five-agent company

Five agents in different roles handled a simulated outage. Their handoffs left out part of the cause. The official record stayed wrong even after all five received the missing evidence. A single agent kept the cause in the record.

How the account changed between roles:

  1. FinanceRelease timing
  2. EngineeringRelease sequence
  3. OperationsMonitoring

The billing clock stayed stopped, even though the outage had reached the threshold for customer credit and review.

A shared memo or earlier decisive evidence prevented the drift in earlier scenarios.

Company experiment and published trace ↗

COUNCILS & MINORITY IDEAS

Idea loss in LLM councils

Choosing one answer kept 49 of 134 valuable ideas raised by only one model, compared with 32 after peer review. For ideas raised by several models, peer review led with 6 of 17.

Across 16 prompts, model judges rated the ideas without knowing where they came from.

In a separate follow-up, recording proposed ideas and checking the draft for omissions preserved more of them.

Council study and comparisons ↗

STANDARDS & COLLECTIVE DELIVERY

Completed modules with incompatible interfaces

Four firms each had to build one module of a shared system and choose compatible interfaces. All four runs completed the modules, but none delivered the final system.

0 / 4 runs delivered the final system.

The formats show one of the four runs.

Standards benchmark and retained results ↗

Firm size and coordination

When does a team of agents beat one agent on the whole job?

Ten agents given a shared specification and automated checks passed all the tests. One frontier agent also passed them, using about a quarter of the tokens and taking over three times as long.

Same complete builds · six runs per arrangement
ArrangementTests passedMean time
Ten, no shared spec60%137 s
Ten, shared spec + checks100%100 s
One frontier agent100%355 s

Both teams used the same worker model. Time includes writing the spec. When requirements changed, old checks reverted valid fixes. Teams allowed to revise the shared code passed the updated tests.

The Coasean Ceiling ↗

MARKETS & SHARED ATTENTION

Agent adoption and provider congestion

In this simulation, agents sent more requests and targeted them better. Providers could answer only so many. When more customers used agents, inboxes filled up and fewer customers found a match.

The market study & its assumptions ↗

Compute allocation

Who gets compute when a GPU cluster fills up?

Twelve firms competed for a fixed pool of compute. The auction raised the overall score while fewer science jobs finished on time. A fixed rule improved both measures over the auction.

48 runs across three simulated shocks
Allocation ruleOverall scoreScience on timeStartups on time
Flat pricing76.840%38%
Spot auction88.229%69%
Fixed rule94.850%74%
TokenFed governor99.667%66%

Overall score combines customer output and lab R&D. Compared with the fixed rule, the governor finished more science jobs on time, but fewer startup jobs.

Economic scenarios

What stays expensive when digital work gets cheaper?

Across three scenarios, cheaper digital services coexisted with rising physical prices. The Assurance scenario assumed insurers and regulators accepted automated software verification, allowing more legacy systems to be upgraded.

Annual price change at H2 2030 · three simulated scenarios
ScenarioConsumer digital CPIPhysical CPI
Baseline−3.60%+2.93%
Automated Assurance−4.80%+2.94%
Credit Crunch−4.20%+2.80%

CPI is the consumer price index. Digital services make up 5% of the simulated basket; physical goods and services make up 95%.

World 2030 ↗

Lessons

If your agents get stuck at a handoff or overload a shared queue, send me the case ↗. I’d like to talk through what we could change and how we’d test it.

Studies, simulations and tools

23PROJECTS

Side project: Evolora, an early artificial-life prototype.