Thesara AI / Applications

The world's core software, rebuilt for the model era.

HR moved from the mainframe to the desktop, and from the desktop to the cloud. Each time it was rewritten, and each time the companies that rewrote it first took the market. It is being rewritten again — this time around the model.

The seven suites → Platform architecture

The argument, in one example

Take hiring. Follow it through three platforms.

On client–server, a recruiter opened an applicant tracking system installed on a company server. The software's job was to hold records and enforce a workflow. A human did all the work; the software remembered it.

On cloud, the same application moved to a vendor's servers and was billed per seat per month. The recruiter got a better interface, mobile access and continuous updates. The software's job was unchanged: it still held records and enforced a workflow. A human still did all the work.

AI-native is a different object. The system reads the market to draft the role, sources and screens the pipeline, runs first-pass structured interviews, schedules the panel, drafts the offer, flags the bias risks in its own shortlist, and writes down why it did each of those things in a form a regulator can inspect. The recruiter's job becomes judgement and exceptions. The screen is no longer the product — the outcome is.

You cannot get from the second to the third by adding a chat box to the second. The data model, the permission model, the audit model and the pricing model are all different. That is why the incumbents are vulnerable, and it is why this is a build rather than a bolt-on.

The market has already priced this in. Gartner estimates up to $234 billion of enterprise application spending is exposed to agentic disruption between now and 2030 — roughly a fifth of enterprise application SaaS spend by that year — as agents complete work across systems instead of people clicking through them (Gartner, July 2026).

$234BEnterprise app spend exposed to agentic disruption by 2030 — Gartner

~20%Share of enterprise app SaaS spend affected by 2030 — Gartner

40%Enterprise apps embedding task-specific agents by end-2026, from under 5% in 2025 — Gartner

>40%Agentic AI projects Gartner expects to be cancelled by 2027 — the reason execution beats enthusiasm

Platform

Build the hard part once.

Every suite sits on the same core. That is the whole economic argument for building seven products instead of one: the seventh costs a fraction of the first, and every improvement to the core lifts all of them at once. The same core is what our services teams deploy on client engagements.

LAYER 5 · INDUSTRY SUITES

Human Resources · Health · Pharmacy · Education · Textile · Finance · Retail

LAYER 4 · EXPERIENCE

Task surfaces, approvals and exception queues for the human in the loop; API and event surfaces for the systems around it.

LAYER 3 · AGENTS & ORCHESTRATION

Planning, tool use, multi-step execution, human escalation rules, and deterministic guardrails around every action that touches money, health or employment.

LAYER 2 · KNOWLEDGE & MODELS

Retrieval over client data, industry-tuned models, evaluation harnesses, and model routing that picks the cheapest model able to meet the quality bar for each task.

LAYER 1 · TRUST

Identity, permissions, tenancy, encryption, data residency policy, full decision logging and replay. Non-negotiable and shared by every suite above.

LAYER 0 · COMPUTE

Established cloud and specialist compute providers today, selected per region for residency and cost. Thesara-operated facilities are a later chapter, not this one.

Residency is inherited

A suite does not implement data residency. It inherits it from Layer 1, which is why a health deployment in Frankfurt and a retail deployment in São Paulo can share a codebase without sharing a jurisdiction.

Every decision is replayable

Inputs, retrieved context, model version, tool calls and outputs are logged for each agent action. An auditor can reconstruct why a system made a specific decision on a specific day — the standard regulated industries will require and most vendors cannot meet.

Model-independent by design

Layer 2 routes to the best available model per task, including open-weight models we host ourselves. That protects margin, protects the customer from a single vendor's pricing, and keeps sensitive workloads where they belong.

The suites

Seven industries, chosen for a reason.

Each of these runs on software written for a previous platform, carries heavy documentation and compliance overhead, and has an obvious outcome to price against. That combination is what makes a rebuild worth doing.

Suite 01 · Human Resources

From record-keeping to hiring, done

Sourcing, structured screening, interview scheduling, onboarding, payroll operations, performance and workforce planning.

Broken today: recruiters spend most of their week on scheduling and screening admin
AI-native: the pipeline is worked continuously; humans handle judgement calls and exceptions
Priced on: roles filled and time-to-hire, alongside a per-employee base
Proof required: bias testing and full decision logs — employment decisions are legally reviewable

Suite 02 · Health

Give clinicians their evenings back

Ambient clinical documentation, coding, scheduling, prior authorisation, referrals and revenue cycle.

Broken today: documentation and billing consume hours of every clinical day
AI-native: the note, the code and the claim are drafted from the encounter and confirmed by the clinician
Priced on: per clinician, with outcome pricing on denial and rework rates
Proof required: in-jurisdiction processing, clinician sign-off on every artefact, no silent autonomy

Suite 03 · Pharmacy

Dispense faster without dispensing wrong

Prescription intake, interaction and allergy checking, inventory and cold chain, adherence outreach, insurance adjudication.

Broken today: phone-and-fax workflows wrapped around a dispensing system built decades ago
AI-native: intake, checking and adjudication run continuously; the pharmacist reviews and authorises
Priced on: per location, plus per adjudicated claim
Proof required: deterministic safety rules that a model cannot override, and a complete audit trail

Suite 04 · Education

One teacher, thirty different learners

Curriculum planning, adaptive tutoring, assessment and feedback, admissions and student administration.

Broken today: instruction is paced for the middle of the room and assessment arrives too late to act on
AI-native: practice adapts per learner in real time; teachers get the diagnosis, not just the score
Priced on: per learner per year, with institutional licensing
Proof required: curriculum boundaries the model must stay inside, and strict protection of minors' data

Suite 05 · Textile

The least digitised large industry on earth

Design and sampling, materials and costing, production planning, quality inspection, and supply-chain traceability to the fibre.

Broken today: spreadsheets, email and physical samples across a supply chain spanning ten countries
AI-native: visual quality inspection, demand-linked production planning, and traceability that survives an audit
Priced on: per production line and per traced unit
Proof required: works where connectivity is poor, and where the shop floor is not a desk

Suite 06 · Finance

Explainable, or it does not ship

Underwriting and credit decisioning, reconciliation, treasury, AML and fraud monitoring, regulatory reporting.

Broken today: reconciliation and compliance reporting still absorb enormous manual effort
AI-native: continuous reconciliation and monitoring, with the exception queue as the human interface
Priced on: per transaction volume band, plus per seat for analysts
Proof required: model governance, challenger models and adverse-action explanations an examiner accepts

Suite 07 · Retail

Decisions at the shelf, not at head office

Demand forecasting, assortment and pricing, replenishment, store operations, customer service and returns.

Broken today: forecasts are weekly and central; the shelf is empty now and locally
AI-native: per-store, per-SKU decisions made continuously rather than in a Monday planning cycle
Priced on: per store per month, with upside sharing on markdown reduction
Proof required: never lets a model set a price outside approved guardrails

Roadmap

Order of build

Wave 1 — Human Resources and Health. The broadest buyer and the hardest compliance bar, deliberately taken together.
Wave 2 — Finance and Retail. Highest transaction volumes; proves the platform under load.
Wave 3 — Pharmacy, Education and Textile, each drawing on components already built.
Wave 4 — Logistics, insurance, legal and public administration, ordered by demand.

The compounding effect

Why products and services belong in the same company.

A pure product company guesses at what enterprises need. A pure services company rebuilds the same thing for every client and owns nothing at the end of it.

Thesara AI runs the loop deliberately: services engagements reveal exactly where real organisations break, those lessons harden the platform, the hardened platform makes the next engagement faster and cheaper to deliver — and the components that keep recurring become the next product.

01 → A client engagement exposes a real workflow problem
02 → The fix is built into the shared platform, not the client fork
03 → The next engagement starts further along and costs less
04 → Recurring patterns graduate into a productised suite
05 → Subscription revenue replaces project revenue
06 → Margin funds the next suite · back to 01

See how we actually deliver.

The services page sets out the engagement model, the delivery method, and what a client can expect in the first ninety days.

Services & delivery → The opportunity