Research · Harare
Jacaranda Labs AI Think Tank
Research on governable autonomy for the smallest firms: how a one-person business can let software act on its behalf without harming strangers, its customers or itself.
Programme
Distributed Cognitive Autonomous Infrastructure
The programme's working claim is that trustworthy autonomy is infrastructure before it is intelligence: credentials, budgets, observers and ledgers. That infrastructure has to run on resources the smallest firm already has. The programme tests the claim on working ventures rather than on demonstrations.
Questions we work on
- Constraint-native design: what a business model looks like when it is designed around quotas, a zero budget and third-party rails.
- Simulated completeness: how leaders tell a working mechanism from a convincing surface, and when to audit and rebuild.
- Unwitnessed autonomy: what autonomous growth systems do to third parties when nobody is watching, and how to bound them.
- Emotional engineering: trust, agency and ethical adoption for the people who operate systems that act.
Publications
Books and working papers
All are written by Courage Garikai. Published papers can be read online; private author-review editions are listed but not yet distributed. Each title links to its full catalogue entry.
- BookPrivate author-review
Emotional Engineering for AI-Oriented Infrastructure Operators
Trust, agency and ethical adoption in systems that act
The book expands the commissioned emotions document into a broader guide for AI-infrastructure operators. It treats AFGR as an operator heuristic, not a validated universal psychological sequence. Research, proposed practices, illustrative examples and STPI operating observations are distinguished. No invented author biography, endorsements, customer outcomes or interviews are claimed.
- BookPrivate author-review
The Tight Bucket
Designing constraint-native digital businesses
The book uses TightBucket, a website "revenue-leak" auditing service built in three days in October 2026 by its owner working with AI coding agents, as a running case. TightBucket had not launched commercially at the time of writing: it had taken no payments, so the book reports design, governance and measured operating data, not commercial…
- BookPrivate author-review
Proof of Mechanism
How leaders who build with AI agents tell a working product from a convincing demo
The book uses AttestOS, a bot-verification service, as a candid running case. Its first version (26 October 2026) presented every conventional sign of completeness while lacking most mechanisms. An audit the next day logged 65 findings, 18 of them critical. The service was rebuilt that evening and checked again on 28 October. The book does…
- BookPrivate author-review
Nobody Was Watching
How founders run autonomous growth systems without harming strangers, their brand or themselves
The book's case is Tapstall, a shop-link service for informal merchants. On the day of its audit, its test suite passed in full while its growth pipeline was emailing non-prospects, posting into another venture's social account and running duplicate copies of itself. Each chapter sets out a question, the theory, the case, a framework, the…
- Working paperPrivate author-review
Constraint-Native Business Model Design
Quota-bounded autonomy, proof-first acquisition and verification debt in a zero-cash, AI-built venture
Generative AI coding tools and generous cloud free tiers have pushed the cash cost of building a working software business towards zero. When building stops being the scarce input, the design problem for a new venture changes. This paper asks what a business model looks like when it is designed around hard external constraints rather…
- Working paperPrivate author-review
Simulated Completeness
Claim–mechanism gaps, mechanism audits and the rebuild decision in an AI-era trust product
Generative AI has made the visible surface of a software product (interfaces, dashboards, documentation, screenshots, test badges, release notes) far cheaper to produce than the mechanisms that make the surface true. This paper names the resulting condition simulated completeness: an artefact whose surface signals operational completeness that its mechanisms do not support. It reports a…
- Working paperPrivate author-review
Unwitnessed Autonomy
Externalities, borrowed identities and duplicate clocks in an autonomous growth pipeline
Small ventures can now run growth pipelines that act without a person in the loop. A scheduled job finds prospects, another writes to them, a third posts to social accounts, and a payment webhook switches accounts on. Each component can be correct by its own tests while the pipeline as a whole does things that…
- Research paperRead online
The Sovereign Micro-Enterprise
A Governance-First Architecture for Zero-Capital Agentic Businesses
Frontier-class language models have pushed the marginal price of machine cognition toward zero, and the standard inference drawn from that collapse is that any connected individual now effectively owns a software company. The inference proves fragile in practice: producing content, code, or conversation is easy, while operating — posting on schedule, selling inside the law,…
- Working paperRead online
Order Without Custody
Governance-first design for merchant commerce rails in low-infrastructure economies
Small merchants already sell at scale — through chat threads, pages and a phone number painted on a wall — but they sell without a record. This paper argues that the binding constraint at that tier is not payment capability but reconciliation, and that reconcilability is an architectural property rather than a feature. Five mechanisms…
- Working paperRead online
Grounded Cadence
Fact-bound content operations as a governance architecture
Anyone can now emit a week of social copy. Almost no one can emit it on a schedule, from a company's own facts, without inventing customers or numbers, without spending the operator's quota on a stranger, and without a silent failure looking like a green run. This paper reports a live zero-capital writing service and…
Earlier works
- The Governable Enterprise: A Practical Guide to Evidence, Authority and AI Operations: Lessons from DeskOS
- From Spreadsheet Interface to Governable Autonomy: DeskOS as a Critical Design-Science Case of Evidence, Agency and Resource-Constrained AI Infrastructure
Standards
How we work
Relationships disclosed
Every work names its case and states that the case belongs to the same owner. Insider access is declared, not hidden.
Claims labelled
We separate what was observed in code and telemetry from what is proposed as a hypothesis or proposition for others to test.
Review before release
Works circulate as private author-review editions first. None has been peer reviewed, and none is presented as if it had.
Failures included
Defects, silent failures and rebuilds are part of the record, because they are where the useful evidence is.
Funding and independence
Who pays for this
The Think Tank is funded by Jacaranda Labs, which is owned by Courage Garikai, the author of its works. Jacaranda Labs also owns every venture used as a case. The research is therefore not independent of its subjects, and it says so.
Contact
Collaborate or request a copy
Researchers, editors and reviewers can write to couragegarikai@atomicmail.io to request a review copy or to propose collaboration.