A no-code AI agent builder that tests every skill against figures your business already signed off
A new data assistant built by an analyst in minutes, with no code.
See one case, screen by screen ↓A tool that lets the people who know the data build an AI agent by describing it, without writing code or a pipeline. In the No-Code Agent Builder a supply chain analyst describes the agent in plain words; the Agent Builder reads the warehouse catalog, proposes tables and joins, writes down the rules and drafts skills, each a question with the query that answers it.
How to choose AI solutions for life sciences →
Updated
The supply chain analyst knows which tables answer which question, how freight is counted and which figures finance trusts. But she cannot write the pipeline, so a new analytics question still becomes a ticket. It waits about a month in the engineering queue for someone who does not know the data. Meanwhile planners and buyers decide mode, carrier and lane on last quarter's spreadsheet.
Asking a general chatbot instead is faster and riskier. One join that repeats spend once per shipment line, or a voided invoice counted next to its re-bill, gives a confident wrong number — and nobody can see the query, let alone test it. In a medical device supply chain, where quarantined lots, field actions and complaint-linked returns sit in the same warehouse, an untested answer is not an answer.
Estimated split for one new analytics question, by request to engineering and with the builder. The 31-day median is the one the solution's dashboard cites; with the builder, the build itself took 14 minutes in the sample, and the rest is the analyst reading and confirming each skill.
Builder agents draft each skill and test it against signed-off figures; the analyst confirms it.
Monday, 10:31. Lena Ortiz, a supply chain analyst at a medical device manufacturer, needs an agent that answers freight cost questions. Here is what happens next, screen by screen, in the working solution.
The Build screen opens on a conversation on the left and an empty agent on the right: purpose, data, rules, skills and the test board. Lena types what people keep asking her — what we spend, with which carrier, per unit on each lane, and how much goes on expedited freight. Beneath, the agents built here before: Returns in 15 minutes, Warehouse capacity in 11, Expiry & short-dated stock in 12, Supplier delivery in 17.
Agent Builder: “I'll find the data, write down your rules, draft its skills and test every one against figures you already trust.”
The builder searches the supply chain warehouse, read-only, and finds 8 tables that look related. It highlights the likely six — freight_invoices (1.1M rows), shipment_lines (96.3M), carriers, sites, customer_accounts and fx_month_end — and asks how they join and which figures she already trusts. On the right the agent now has a name and three questions it will answer.
Lena names the keys, says expedited means air or courier, and points to the September freight report and the Q3 lane review. The builder checks all 5 join keys against the warehouse, writes down 4 rules and takes 9 test questions from the September and August freight reports — month-end close, signed by Omar Haddad — and the Q3 lane cost review.
“A shipment has many lines, so I sum lines per shipment first — otherwise spend would repeat once per line.”
One click drafts three skills — spend by carrier, cost per unit by lane, expedited spend — each a question with the query that answers it, and runs all nine tests live and read-only. Seven pass. September spend comes out at $2,412,300 against $2,291,800 in the report, and the expedited share at 17.1% against 18.0%. The Skill Tester finds one cause for both.
“The $120,500 gap is 118 voided Kestrand Freight invoices that were re-billed in September — counted twice.”
“Yes — leave out voided invoices (status VOID). They were re-billed.” The builder adds the rule to all three skills and re-runs every test: nine of nine pass, September is $2,291,800 and the expedited share 18.0%, matching the report. Each skill shows its test questions, expected figures, sources and answers; a failing test would block Confirm. Lena confirms all three, each stamped with her name and the time.
The Skill Tester never changes an expected figure and never marks a failing test as passed.
Go live saves version 1.0 — instructions, 5 rules, 3 verified skills, 9 tests — and adds the agent to the supply chain assistant's routing. Three probe questions route as expected: Kestrand spend (0.94) and air freight share (0.91) to Freight cost, and the least on-time carrier (0.89) stays with Shipments & carriers.
“Freight cost agent is live, built in 14 minutes.”
Lena asks the way her colleagues will: what did a unit cost to move from Leiden to Madrid in September? The assistant routes it to Freight cost (match 0.95). $0.84 per unit — $351,900 of freight across 418,900 units on 1,912 shipments — split by carrier, with a note that courier is 10% of units but 24% of the lane's spend.
“See the query” opens the exact query that ran — read-only, 1.2 s, 46 lanes — with the lines that produced the figure highlighted, including the new VOID rule. Below it, the skill's three test questions and their sources. Customer contact columns are hidden from every answer.
“Highlighted: the lines that produced the cited number.”
Freight cost now has its own page: 6 tables and 5 join keys, the rules it follows from the build conversation, a Try it box, and versions — v0.1 draft at 10:31, v1.0 live with 3 skills confirmed and 9 of 9 tests passing. Each night its tests run again.
The same freight cost agent, seen by the four people around it — what a new question cost them before, and what it costs now.
Builder agents draft each skill and test it against signed-off figures; the analyst confirms it.
Interviews the analyst in plain words, reads the warehouse catalog, captures tables, join keys and rules, drafts 3–5 skills as question-and-query pairs and takes at least three test questions per skill from reports the analyst trusts.
Runs every test question of every drafted skill live and read-only, compares each answer with the expected figure and its source, and explains the cause of a mismatch in one sentence — voided or duplicate records, fan-out from joins, missing currency conversion, empty periods.
Receives every question, matches it to the agents by purpose and verified example questions, and routes it to the best match — or asks the person which area they mean.
Confirms every skill before it goes live. The agents propose; a named person decides.
Ask about any agent, skill or test in plain words — or tell the assistant what to change. Changes are recorded as yours.
Why did two freight tests fail?
Both failures had one cause. September spend came out at $2,412,300 against $2,291,800 in the September freight report: 118 voided Kestrand Freight invoices ($120,500) were re-billed in September and counted twice. The same error moved the expedited share from 18.0% to 17.1%. Leaving out voided invoices fixes both.
Which agents answered the most questions last week?
Inventory & stock positions answered 412 questions, then Backorders with 288 and Shipments & carriers with 231. Together they took 67% of the 1,395 questions answered from Sep 28 to Oct 4.
How does the assistant decide which agent answers?
Every question goes to the supply chain assistant first. It compares the question with each agent's purpose and verified example questions, and routes to the best match when it is at least 0.80. Below that it asks the person which area they mean. A new agent joins the routing the moment you confirm it — there is no deployment step.
Add a rule: never show customer contact names in an answer
Added to Rules every agent follows. It applies to every live agent from the next question, and it is recorded in the audit trail as a change by you.
13 screens from the working solution, on its sample data. Pick one to see it large.
Who confirms a skill, how strict the tests are and what the agents may read are settings, changed in one place and recorded in the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Who confirms a new skill | The analyst who builds it | The analyst who builds it · The analyst and a data steward · A data steward only |
| Skills that read cost or price tables also need | Omar Haddad · Data steward | Omar Haddad · Data steward · Dana Okafor · Supply chain lead |
| Test questions per skill, at least | 3 | 1 · 3 · 5 |
| A skill can be confirmed only when every test passes | On | On · Off |
| A figure matches when it is within | 0.1% | 0% · 0.1% · 0.5% |
| Ask for a re-check when a result moves more than | 10% | 5% · 10% · 20% |
| Most rows a query may return | 50,000 | 10,000 · 50,000 · 250,000 |
| Route a question when the match is at least | 0.80 | 0.70 · 0.80 · 0.90 |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate of where the days go · The 31-day median wait is the figure the solution's own dashboard cites for requests before the builder. People, companies, carriers, sites and figures named on this page are fictional — characters and sample data in the working solution.
A tool that lets the people who know the data build an AI agent by describing it, without writing code or a pipeline. In the No-Code Agent Builder a supply chain analyst describes the agent in plain words; the Agent Builder reads the warehouse catalog, proposes tables and joins, writes down the rules and drafts skills, each a question with the query that answers it.
Every skill must pass test questions taken from reports the business has already signed off, such as the month-end freight report or the quarterly lane review. The Skill Tester runs each test live, shows pass or fail with the expected figure and its source, and explains the cause of any mismatch. A skill with a failing test cannot be confirmed.
No. The connection is read-only: SELECT statements only, a 50,000-row cap and a 30-second timeout. The agents never write to the warehouse, and customer contact details are hidden from every answer.
Yes. Every answer names the agent and the verified skill that produced it, and one click opens the exact query that ran, with the lines behind the figure highlighted and the skill's test questions below.
No deployment step. Once the analyst confirms the skills, Go live saves the version and adds the agent to the supply chain assistant's routing straight away. In the sample build, the freight cost agent was live 14 minutes after the first message.
The Skill watcher re-runs every live skill's tests each night. When a result moves more than the set limit — 10% by default — it asks the owner to re-check the skill before trusting it again.
Yes. No skill reaches an agent until it passes its tests and the person who knows the data confirms it. Skills on cost or price tables also need the finance data steward, every version is kept, and every step is on the audit trail.
The Agentic Solution Engine builds and deploys the No-Code Agent Builder from your requirements and documents — your warehouse catalog, the reports you sign off and your sign-off rules — and it goes live once every quality gate has passed. We will build a first agent with your analysts on your own data.
Any backorder or stock question answered in seconds, from the source systems.
1-min demoDrug Shortage Early WarningRanked, cited shortage alerts raised weeks before stock runs out.
1-min demoTemperature Excursion ManagementEvery temperature alarm turned into a cited release-or-reject call in hours.
1-min demoWe'll build a first agent with your analysts, tested against a report your business has already signed off.