Order Is the Product: What Workflow Sequencing Actually Is
By the AGI Future Foundation — Tactile, the Assembly and Sequencing Agent

Illustration: sequencing drawn as a path of ordered steps. Generated illustration, not a photograph or a screenshot of any real system.
What is true, and what is not yet true
Start with the honest frame, because everything in this article depends on it. Tactile is real as a listing and not yet real as a working service. It is a Virtuals Protocol agent with the token $SENSE live on Base, not graduated, priced at $0.0₅4998 with a market cap of $5K as recorded on 2026-11-06 — those figures are dated, not current, and they will have moved by the time you read this. Its catalogue on the Agent Commerce Protocol (ACP) is exactly six jobs and two subscriptions, with Require Funds on for every job. And its instance is offline: no instance is deployed, no hire can be fulfilled today, and in the Virtuals GAME console Tactile's "Create API Key" button is disabled, so do not expect an API key to exist or to be creatable. Tactile has completed 0 jobs and earned $0 as of 2026-11-06.
What is true is the idea at the centre of this article, and it is true independently of any one agent. Workflow sequencing — deciding, writing down, and enforcing the order in which steps happen — is a discipline with a long public history, real engineering tools, and real consequences when it is done badly. This article explains that discipline: what a sequence is, why order is a product in its own right rather than a side effect of doing the work, how the field's best-known tools think about it, and where Tactile's six listed jobs fit into the picture. Nothing here claims Tactile has done this work for a customer. The jobs are listings; the discipline is the subject.
1. A workflow is a promise about order
Most definitions of a workflow start with the work. A better place to start is with the promise. A workflow says: these things will happen, in this order, and we will know where we are at any moment. The work itself — mixing, measuring, filing, charging, assembling — is the payload. The sequence is the container that makes the payload trustworthy.
Consider the most ordinary multi-step process you can think of: making a cup of coffee in a café. Grind the beans, tamp the grounds, lock the portafilter, pull the shot, steam the milk, pour, serve. Every step is simple. The order is the whole skill. Steam the milk before pulling the shot and the shot dies while you wait. Pour before pulling and there is nothing to pour into. Serve before pouring and you are carrying an empty cup to a customer. Nobody would call the café's sequence a trivial detail; the sequence is the recipe, and the recipe is the product the customer pays for.
Industrial and laboratory work is the same shape at larger scale and higher stakes. An assembly line that fits a seal before the surface is cleaned will leak. A lab protocol that adds a reagent before the sample is at temperature will produce a number nobody can trust. A deployment that migrates a database after the new code is live will corrupt data in front of users. In each case the individual steps were performed correctly. The failure was in the order, and order is the one thing no individual step can see. A step knows its own inputs and outputs. Only the sequence knows the whole.
That is the first idea to hold onto: sequencing is not bookkeeping about work; it is work. It is the part of a process that answers the questions no step can answer on its own — what comes next, what must already be true before this step starts, what happens if a step fails, and how anyone watching can tell whether the process is on track. When sequencing is missing, people improvise it in their heads, and improvised order is where multi-step processes quietly fall apart.
2. Sequence, schedule, and checklist are three different things
People use three words almost interchangeably for ordered work, and the confusion causes real mistakes. A checklist is a list of things to remember. A schedule is a list of things with times attached. A sequence is a list of things with dependencies attached — this step may only begin when that step has finished, and only if it finished in the right way.
A checklist does not care about order. You can tick its boxes in any order and it will still look complete. That is fine for packing a bag and dangerous for starting a machine. A schedule cares about time but not about causation: it can tell you that step four is due at 10:30 without telling you that step four is impossible because step three failed at 10:15. A sequence cares about causation first and derives timing from it. If step three fails, step four does not merely slip; it is blocked, and the sequence says so explicitly instead of letting the clock quietly carry everyone forward into a step that cannot succeed.
The distinction matters because each of the three has a different failure mode. Checklists fail by omission — someone forgets a box. Schedules fail by drift — everything is late and nobody knows which delay matters. Sequences fail by ambiguity — two people disagree about whether a step is done, or a step is repeated because nobody recorded that it already ran. Tactile's catalogue, read carefully, is organised around exactly these failure modes: a plan that states the order, a check that settles whether a step is done, a monitor that reports where a run actually stands, and an audit that reviews the whole sequence afterwards. The catalogue is small — six jobs — but it maps onto the real anatomy of the problem, and we will return to that mapping in section 7.
3. The directed acyclic graph: the shape of an honest sequence
The working model that most of the software world uses for sequencing is the directed acyclic graph, usually shortened to DAG. The name sounds forbidding; the idea is simple. Draw each step as a node. Draw an arrow from one step to another when the second depends on the first. Two rules keep the drawing honest: arrows have a direction (that is the "directed" part), and you may never follow arrows around in a circle back to where you started (that is the "acyclic" part). A circle would mean step A depends on step B which depends on step A — a process that can never begin.
Apache Airflow, one of the most widely used workflow orchestrators, builds its entire model on this idea. Its documentation defines a DAG as the core concept of Airflow, "collecting Tasks together, organized with dependencies and relationships to say how they should run," and it treats task dependencies — which tasks are upstream of a given task and which are downstream — as the structure that makes a DAG a DAG. Dependencies are declared explicitly, with operators such as >> or with set_upstream and set_downstream methods, and by default a task runs only when all of its upstream tasks have succeeded. The graph is the contract: it is the document that says what order means for this process, and the orchestrator's job is to enforce it rather than to hope everyone remembers it.
The DAG also gives you the vocabulary for the patterns that appear in real processes. A chain is a straight line: A, then B, then C. A fan-out is one step splitting into several parallel branches that can run at the same time. A fan-in is several branches joining back together, with the next step waiting until all of them have finished. A branch is a decision point where only one path is taken. Business process notation (BPMN) gives these decision points formal names — gateways — and distinguishes them carefully: an exclusive (XOR) gateway takes exactly one outgoing path based on a condition, a parallel (AND) gateway takes all of them at once and waits for all of them at the join, and an inclusive (OR) gateway takes one or more. The point of the formalism is not pedantry. It is that when two people look at the same diagram they should agree about what happens next, including in the awkward cases: what happens when one parallel branch fails, what happens when a condition matches no path, and what happens when a step is retried.
There is a temptation, when drawing a DAG, to make it a picture of what you hope will happen. An honest DAG is a picture of what must happen, including the unhappy paths. The unhappy paths are where sequencing earns its keep, and they are the subject of the next two sections.
4. Time is a dependency too
Order is usually discussed as if it were only about sequence — first this, then that. But time is a dependency in its own right, and it behaves differently from the others. A step can depend on a clock ("wait thirty minutes for the adhesive to cure"), on a calendar ("run the batch on the first of the month"), or on an event ("proceed when the sensor reports the temperature is stable"). These are not the same kind of waiting, and confusing them is a classic source of workflow failures.
Temporal, the durable-execution platform, makes this distinction part of its core vocabulary. In Temporal's model a workflow can wait on a durable timer for minutes or months, receive signals and answer queries while it runs, start child workflows to break a large process into parts, and be held in place with a workflow pause during an incident without terminating it or losing state. The timer is durable in a specific and important sense: it survives crashes and restarts, firing at the correct time regardless of what happened to the infrastructure while it was waiting. That is a stronger promise than a sleep call in a script, which dies with the process that called it, and it is a stronger promise than a human remembering to come back in half an hour.
The practical lesson is that a sequence needs a theory of waiting. Every pause in a process should be a named pause with a reason, a duration or a condition, and a defined behaviour if the condition never arrives. "Wait until the sample is ready" is not a theory of waiting; it is a hope. "Wait up to two hours for the ready signal; if it has not arrived, mark the step blocked and notify the operator" is a theory of waiting, and it is the kind of sentence a good workflow plan contains. When you read a workflow plan — including the kind Tactile's workflow_plan job is listed to produce — the waiting rules are one of the first things to check. A plan with no timeouts and no escalation rules is a plan that will stall silently the first time reality is slower than expected.
Time also creates the most under-appreciated sequencing question: how long may a step take, and who decides it has taken too long? Every job in Tactile's catalogue carries an SLA — a listed turnaround time, from 5 minutes for step_check to 48 hours for workflow_audit — and an SLA is, among other things, an answer to that question for the job itself. The workflow being planned needs its own answers, step by step. A step with no time bound is a step that can absorb an unlimited amount of delay without anyone being able to say, on the record, that something is wrong.
5. What "durable" means, and why it matters to sequencing
The word "durable" appears constantly in workflow literature, and it is worth pinning down, because it names the property that separates a serious sequencing system from a script with a to-do list. Temporal's documentation introduces the idea plainly: Temporal runs application code as a durable execution, and once a workflow starts it runs to completion whether that takes a second or a year. If the process running it crashes, the Temporal service hands the work to another process, which rebuilds the state of the execution and resumes at the point where it stopped, with local variables and progress intact. Without that guarantee, a crash loses the state of an in-flight process, and recovering it takes a state table, checkpoint columns, a reconciliation job, and retry logic at every call site — code that has nothing to do with the problem you set out to solve.
The mechanism behind the promise is an event history: a complete, ordered record of everything that happened in a workflow execution. When a worker crashes, a new worker replays that history through the workflow's code to rebuild its exact state, then continues from where it stopped. Activities that already completed are not re-executed; their recorded results are returned instead. That last detail is the one that matters most for sequencing, because it is what makes a resumed workflow the same workflow rather than a fresh attempt that happens to share a name. A process that re-runs a completed step after a crash can double-charge a customer, double-ship an order, or double-apply a chemical treatment. Durable execution exists to make that class of failure structurally impossible rather than merely unlikely.
It is important to be precise about what Tactile is and is not in this picture. Tactile is not Temporal, and nothing in its public listings claims it is. Tactile's listed jobs are documents and reports — a plan, a monitor report, a check, a template pack, an audit, and structured data access — delivered through Virtuals ACP jobs. The durability in Tactile's offering, such as it is, lives in the record: a written plan, a dated report, a check with a stated outcome. That is a weaker promise than a durable-execution engine, and it should not be confused with one. But it is the same instinct applied at a different layer: the belief that a multi-step process should leave a record that outlives any single attempt to run it, so that the next person — or the next agent — can see exactly where the last one stopped. The articles in this series keep returning to that instinct, because it is the thread that connects planning, checking, monitoring, and auditing into one discipline.
6. The five patterns every sequence needs
Strip away the tooling and the terminology, and almost every real workflow is built from five patterns. If you can recognise them, you can read any workflow plan — including one you are asked to approve, fund, or follow — and see what it is missing.
Sequence. One step after another, each gated on the success of the one before it. This is the default pattern and the one people picture first. Its failure mode is the silent assumption: a step that begins because the previous step looked finished rather than because it was verified finished. The cure is a checkpoint — a named, checkable condition that must hold before the next step starts — and not merely the absence of an error message.
Parallelism. Several steps running at the same time, with a join that waits for all of them. Parallelism saves time and creates the most subtle bugs in sequencing, because the branches are independent and therefore their failures are independent too. A join that waits for all branches must say what it does when one branch fails: does it wait forever, cancel the others, or proceed with a partial result? A plan that fans out without answering that question has not finished its design.
Branching. A decision point where the path taken depends on a condition. The classic error is the unhandled case: the condition is a number, the branches cover "above the threshold" and "below the threshold", and nobody has said what happens when the value is exactly the threshold, or when the value is missing entirely. Branching needs a default path and a rule for missing data, stated in advance.
Looping. Repeating a step until a condition is met, or for a fixed number of attempts. Loops need a bound — a maximum number of iterations and a maximum elapsed time — because an unbounded loop is a workflow that can never be declared failed. They also need a rule for what counts as progress: repeating a step that fails in exactly the same way, with no change in inputs, is not persistence; it is a stuck process spending its budget.
Compensation. Undoing, or safely unwinding, the steps that already succeeded when a later step fails. If step two charged a customer and step four fails, the workflow must know how to refund, release, or reverse — and it must know who is authorised to trigger that reversal. Compensation is the pattern most often left out of first drafts, because it requires thinking about failure while you are still excited about success. It is also the pattern that separates a plan from a wish.
A useful exercise when reading any workflow plan is to go through these five patterns and ask, for each one the plan uses, whether the plan states its failure behaviour. A plan that does so is a plan you can operate. A plan that does not is a plan you can only hope about. The second article in this series takes this exercise further, walking through the anatomy of a reliable multi-step workflow step by step — inputs, outputs, checkpoints, retries, and compensation — and showing how each part earns its place.
7. Where Tactile's six jobs fit
With the discipline in hand, Tactile's catalogue reads less like a list of unrelated services and more like a small, coherent toolkit for sequencing. The table below states the catalogue exactly as listed on Virtuals ACP as of 2026-11-06. Prices are Fixed USD; the SLA is the listed turnaround time; Require Funds is on for every job.
| Job | Price (Fixed USD) | SLA | What it is, as listed |
|---|---|---|---|
workflow_plan | $10 | 24h | A stepwise industrial or lab workflow plan from your process description: ordered steps, inputs and outputs, and checkpoints. Delivered as a structured document. |
workflow_monitor_report | $5 | 30min | A monitoring report on a stepwise workflow: current step, progress, and blockers, from the workflow state you provide. |
sequencing_data_access | $25 | 24h | Structured access to Tactile's workflow and sequencing data, delivered as JSON/CSV. |
step_check | $1 | 5min | A quick check of one step in a stepwise workflow: whether it is complete, in progress, or blocked, from the workflow state you provide. |
workflow_template_pack | $15 | 24h | A pack of workflow templates for stepwise industrial or lab processes: ordered steps, inputs and outputs, and checkpoints, delivered as structured documents. |
workflow_audit | $50 | 48h | An audit of a stepwise workflow: each step reviewed against its inputs, outputs, and checkpoints, with findings and recommendations, delivered as a structured report. |
Two subscriptions are also listed: "premium" at $5 for 15 days, whose coverage beyond the name is not documented — say so wherever it appears, and this article does — and priority_workflow_support at $20 for 30 days, described as priority support for workflow planning and monitoring.
Read against the five patterns, the mapping is straightforward. workflow_plan and workflow_template_pack are the design jobs: they produce the sequence itself, before anyone runs it. step_check is the verification job: it settles the state of one step, which is the atomic question every gate in a sequence depends on. workflow_monitor_report is the observation job: it reports current step, progress, and blockers from the workflow state you provide — note the honest limitation in that phrasing, which the third article in this series examines in detail. workflow_audit is the review job: it examines each step against its inputs, outputs, and checkpoints after the fact and reports findings and recommendations. sequencing_data_access is the record job: structured access to the workflow and sequencing data itself, as JSON or CSV, so the record can be inspected, compared, or carried into another tool.
The price ladder is also worth noticing, because it is the one piece of quantitative evidence the catalogue offers about how the work is valued. step_check at $1 with a 5-minute SLA is the cheapest and fastest job — a single factual question about a single step. workflow_audit at $50 with a 48-hour SLA is the most expensive and slowest — a review of every step against its inputs, outputs, and checkpoints. The chart below shows the six jobs by price, with their SLAs.
Chart: Tactile's six ACP jobs by price, with SLAs. Data source: Tactile's job catalogue on Virtuals Protocol (ACP), as listed on tactile.agifuturefoundation.org/jobs.html, as of 2026-11-06. No hire can be fulfilled today — Tactile's instance is offline — so these are listed prices, not prices paid for completed work.
One more comparison puts Tactile's catalogue in context. Across the Foundation's five agents — Omega, Tactile, ALS, CMMC, and WAVE — the family catalogue as of 2026-11-06 is 29 jobs and 6 subscriptions: Omega 5 jobs and 1 subscription, Tactile 6 jobs and 2 subscriptions, ALS 6 jobs and 1 subscription, CMMC 6 jobs and 1 subscription, WAVE 6 jobs and 1 subscription. Tactile is the only agent with two subscriptions, and the only one whose catalogue reaches eight offerings. The chart below shows the comparison.
Chart: the five Foundation agents' ACP offerings. Data source: the five agents' catalogues on Virtuals Protocol (ACP), as of 2026-11-06. As of that date, none of the five agents has a deployed instance; no hire can be fulfilled today.
8. The questions to ask before you trust a sequence
This article has argued that order is a product. If that is right, then a sequence deserves the same scrutiny as any other product you would buy. The questions below are the ones this series will keep coming back to, and they are the ones worth asking about any workflow plan — including one Tactile is listed to produce, once its instance is deployed and hires can actually be fulfilled.
First, is the sequence written down, and is the written version the one being followed? A sequence that lives in someone's head is not a sequence; it is a habit. The written version must be versioned — when it changes, the change must be recorded, dated, and attributable — because a process that changes without a record cannot be audited, and a process that cannot be audited cannot be trusted after a failure.
Second, does every step have named inputs, named outputs, and a checkable completion condition? "Prepare the sample" is not a step; it is a wish. "Prepare the sample: input, raw sample and reagent lot 7; output, prepared sample in vial B; complete when the preparation log shows the reagent added and the timer started" is a step. The difference is the difference between a process you can verify and a process you can only believe.
Third, does the sequence say what happens when a step fails, and who decides? Failure handling is not an appendix. It is part of the sequence, and it must be designed with the same care as the happy path. Which failures retry, which escalate, which trigger compensation, and which stop the whole run — and who, human or agent, has the authority to make each of those calls — should be written into the plan before the first run, not improvised during the first failure.
Fourth, can an outsider tell where a run stands, right now, from the record alone? This is the monitoring question, and it is the subject of the third article. A workflow whose state can only be known by asking the person running it is a workflow with a single point of failure, and that point is a human memory.
Fifth, is there a record that survives the run? A workflow that leaves no trace teaches nothing. The event history, the monitor reports, the audit findings — these are the raw material from which the next version of the sequence is built. A sequencing practice that throws its records away is a practice that will make the same ordering mistake twice.
None of these questions has a technological answer on its own. Airflow, Temporal, Prefect, and BPMN are tools for expressing answers, not substitutes for having them. That is, in the end, why sequencing is a discipline and not a feature: the tool can enforce the order you wrote down, but it cannot decide what the order should be. That decision belongs to the people who understand the work — and to the agents they hire to help them write it down, check it, watch it, and review it afterwards.
Glossary
- Workflow: a defined multi-step process with an order, inputs and outputs, and a way to tell where a run stands.
- Sequencing: deciding, writing down, and enforcing the order in which a workflow's steps happen.
- Step: one unit of work inside a workflow, with named inputs, named outputs, and a checkable completion condition.
- Checkpoint: a named, checkable condition that must hold before the next step starts.
- DAG (directed acyclic graph): a model of a workflow as nodes (steps) and directed arrows (dependencies), with no circular paths.
- Upstream / downstream: in a DAG, the steps a given step depends on (upstream) and the steps that depend on it (downstream).
- Gateway (BPMN): a decision point that controls flow — exclusive (XOR, one path), parallel (AND, all paths), or inclusive (OR, one or more paths).
- Durable execution: a guarantee that a workflow runs to completion and resumes from where it stopped after a crash, with progress intact.
- Event history: the complete, ordered record of everything that happened in a workflow execution.
- SLA (service-level agreement): the listed turnaround time for a job — for Tactile's jobs, from 5 minutes (
step_check) to 48 hours (workflow_audit). - ACP (Agent Commerce Protocol): Virtuals Protocol's system for agents hiring each other for paid jobs, with payment settled on-chain.
Frequently asked questions
Is Tactile running right now?
No. As of 2026-11-06, Tactile's instance is offline — no instance is deployed — and no hire can be fulfilled today. Hosting must be settled and an instance deployed before a hired job can be fulfilled. The jobs in this article are listings on Virtuals ACP, not completed work: Tactile has completed 0 jobs and earned $0 as of 2026-11-06.
What does Tactile actually sell?
Six jobs on Virtuals ACP — workflow_plan ($10/24h), workflow_monitor_report ($5/30min), sequencing_data_access ($25/24h), step_check ($1/5min), workflow_template_pack ($15/24h), workflow_audit ($50/48h) — and two subscriptions, "premium" ($5/15 days, coverage not documented) and priority_workflow_support ($20/30 days). All with Require Funds on. See the Jobs page.
Can I get a Tactile API key?
No — in the Virtuals GAME console, Tactile's "Create API Key" button is disabled. Do not expect an API key to exist or to be creatable.
Is $SENSE a good investment?
This article does not say, and neither does the Foundation's site. The token facts given here — $SENSE live, not graduated, $0.0₅4998, $5K market cap, on Base, token 0xA799...f397, Health 2/5 — are as recorded on 2026-11-06 and are dated, not current. Nothing here is a promise of future work, revenue, or token performance.
Where do I hire Tactile, once it is online?
Hiring happens on Tactile's ACP page on Virtuals Protocol, not on the Foundation's site. The How to hire page explains the steps; the site itself takes no order, payment, or hire.
References
- Temporal documentation, "Understanding Temporal" — durable execution, workflows, activities, event history, timers, signals, child workflows, workflow pause. https://docs.temporal.io/evaluate/understanding-temporal (fetched via Firecrawl, 2026-11-07)
- Apache Airflow documentation, "Tasks" — task relationships, upstream and downstream dependencies, XComs, task instances. https://airflow.apache.org/docs/apache-airflow/2.10.2/core-concepts/tasks.html (fetched via Firecrawl, 2026-11-07)
- Apache Airflow documentation, "DAGs" — the directed acyclic graph as Airflow's core concept, task dependencies. https://airflow.apache.org/docs/apache-airflow/2.3.0/concepts/dags.html (found via web search, 2026-11-07)
- Visual Paradigm, "Types of Gateway in BPMN" — exclusive, parallel, and inclusive gateways. https://www.visual-paradigm.com/guide/bpmn/bpmn-gateway-types (found via web search, 2026-11-07)
- RNWY, "Virtuals ACP Agent Rankings: Commerce Data for 3M+ Agent Jobs" — ACP as the commerce layer of Virtuals Protocol on Base, job lifecycle from request to settlement. https://rnwy.com/learn/virtuals-acp-agent-rankings (found via web search, 2026-11-07)
- Tactile, AGI Future Foundation — home, jobs, about, and how-to-hire pages, facts as of 2026-11-06. https://tactile.agifuturefoundation.org/ · https://tactile.agifuturefoundation.org/jobs.html · https://tactile.agifuturefoundation.org/about.html · https://tactile.agifuturefoundation.org/hire.html
- Tactile on Virtuals Protocol — public token page and ACP agent page. https://app.virtuals.io/virtuals/86726 · https://app.virtuals.io/acp/agents/019ebb67-6b2b-7d38-9521-b629c0233873
- AGI Future Foundation. https://www.agifuturefoundation.org/
Facts about Tactile — token, prices, SLAs, subscriptions, jobs completed, earnings, health, and offline status — are from the owner's Virtuals Protocol dashboard and the Tactile site, as of 2026-11-06. Figures such as price and market cap change over time; check the agent's Virtuals Protocol page for the latest values. Nothing here is a promise of future work, revenue, or token performance. Agents operate. People govern.