🧭 Your platform gets quieter when it works


Oct 4, 2026 | Read time: 11 mins.

Kevin Donovan · Architects' Hub
​
Forwarded this email? Subscribe to Architects’ Hub for the next issue.

The quieter it gets

Hey Reader,

You turn on the tap. Water comes out.

You plug something in. It works.

There’s a great deal of engineering, maintenance, and judgment behind both. Most days, you don’t need to think about any of it.

I’ve been thinking about that this week. As a platform becomes more effective, the hum of the machine can get quieter. The noise recedes. Fewer tickets. Fewer messages asking who can approve something. Fewer people needed to get a routine piece of work moving.

The work behind it may be substantial. The experience of using it becomes ordinary.

That’s a useful ambition for architecture. It’s also surprisingly difficult to report.

If your contribution is measured by how many requests reach your team, removing the need for those requests can make progress look like inactivity.

This week we looked at self-service portals, the humans hidden behind them, and the measures that help tell the difference. Today I want to put the whole test in your hands, including where it should stop.

In this edition

  • What a quieter platform should still make visible.
  • The full 4-part Paved Road Test and its scoring.
  • 3 worked paths, with different constraints.
  • A worksheet to use with your team.
  • Community Corner, and the restored 5-Field Guardrail download.

Let’s get into it.👇

________

🧭

________

​

The work behind the ordinary

Last week, Henry Luija shared the newsletter and added an analogy I really liked: the architect as part of a road crew.

Listen to the people using the roads. Fix the potholes. Improve the existing routes. Build new ones as the operating model and business strategy change.

He took the guardrails idea and gave it somewhere to go. That helped shape this week’s discussion.

A road crew’s contribution isn’t fully captured by the number of drivers who stop to ask for help. Some of its best work means they can keep going.

An internal platform has a similar problem. A team that can create an environment, connect an approved data source, or recover from a routine failure without chasing 3 people has gained something real.

But a quieter support channel alone doesn’t prove it. People might have given up, built a workaround, or stopped using the service.

Keep the experience quiet. Keep the evidence visible.

Put request volume beside time to a working result, required human handoffs, and the share of routine requests teams complete without help. Include failures and rework. Check who is using the path and who has left it.

That gives the falling ticket count some context.

________

🧭

________

​

The Paved Road Test, in full

Choose 1 frequently used internal path. Keep it specific: a standard development environment, an approved data feed, or temporary access to a particular production service.

Define who is using it and what “working” means before you start. A submitted form isn’t a working environment. Access granted isn’t necessarily access that lets someone finish the task.

Walk the path with someone who uses it. Use the last 10 real requests where records are available. A 20-minute first pass can reveal the gaps; it won’t manufacture evidence you haven’t collected.

1. Count the required human handoffs

How many people, other than the requester, must act before the result is usable?

Count the approver, the person who configures it, and the person who only forwards the ticket. Record repeated touches too. One person asked 4 times is still a dependency worth seeing.

Beside each handoff, write what it contributes: judgment, execution, or routing. Some will do more than 1.

“Checks whether this new use of customer data is appropriate” is different from “copies the approved values into another system.”

2. Measure elapsed time from ask to working

For those same requests, record the time from the initial ask to a usable result. Include the time between teams.

Use the median, and show the slowest completed case beside it. Keep failed, abandoned, and still-open requests visible rather than quietly dropping them from the account.

If you don’t have timestamps, mark the estimate as an estimate. That missing evidence is a useful finding.

3. Find the furthest point a team reaches alone

Where does the documented path end and the message to a helpful colleague begin?

Try it with someone who doesn’t know the shortcuts. Record the exact step where they need help, and whether that help is required by the process or needed because the instructions don’t work.

For the sample, count how many routine requests reached a working result without help. Give both the count and the denominator.

4. Try the 2 a.m. question

If someone needs this path outside support hours, what happens?

Does it complete? Does it wait until morning? If it fails halfway through, is there a clear recovery route and someone responsible for it?

You can inspect the runbook and test safely during the day. Nobody needs an actual 2 a.m. experiment.😉

For a service intended only for business hours, a clear waiting time may be entirely reasonable. For an emergency recovery path, the same answer could be a serious gap.

Read the score

Thursday’s shorthand was:

A paved road​
0 required human handoffs, minutes to hours.

Check: Can teams finish and recover within its stated scope?

An honest queue​
1 required human handoff, under a day.

Check: Is the judgment useful, and is the wait explicit?

A gate wearing a portal​
2+ required human handoffs, multiple days.

Check: Which stops add judgment, and which repeat work?

Those are diagnostic patterns, not a universal maturity scale. A 2-person review completed in 15 minutes doesn’t fit the third row. A fully automated process taking 3 days isn’t necessarily a fast path.

Keep all 4 findings together. Don’t average them into a reassuring number.

The first 2 give you the shorthand. The independence boundary and the 2 a.m. result tell you where that shorthand breaks down. Zero approvals with a required Slack rescue is still a dependency. An explicit human review can be the right design for a contested decision.

________

🧭

________

​

3 paths, 3 different answers

These are illustrative examples, not client results. The times make the differences concrete.

A standard development environment

A team chooses an approved template. Identity, network boundaries, cost limits, and an expiry date come with it. Automated checks run, the environment starts, and the team can deploy a sample service.

The test: 0 required handoffs. A median of 20 minutes in our example. Teams can complete the standard path alone. At 2 a.m., it still runs; a failed build gives a usable error and a safe retry route.

The score: a paved road, within a defined boundary.

The judgment was made earlier: which workloads qualify, what limits apply, and who maintains the template. A request outside that scope has a separate route. Nobody has decided that every possible environment is safe to create automatically.

A new use of an existing data feed

A team wants to use an established customer feed for a new purpose. The connection is easy. Whether the existing permission covers that use needs a decision.

One accountable data owner reviews the purpose. Once it is approved, provisioning and checks happen automatically.

The test: 1 required handoff. A median of 6 elapsed hours in our example. The team can prepare and validate the request alone, then stops at the purpose decision. At 2 a.m., it waits for the published review window.

The score: an honest queue. State its service hours and show overnight and weekend cases too. A median under a day doesn’t promise that every request finishes within a day.

The next improvement might be a clearer decision brief or an agreed class of already-approved uses. Removing the reviewer before that boundary is settled would leave the decision unresolved.

Temporary production access

The portal asks for a service, a role, a reason, and a duration. Then it creates a ticket.

A manager approves it. A service owner checks it. An administrator grants access. The audit record exists, but each person repeats part of the same check.

The test: 3 required handoffs. A median of 4 days. The team reaches the ticket confirmation alone, then waits. At 2 a.m., the routine path stalls, and the emergency route depends on knowing whom to call.

The score: a gate wearing a portal, with a separate recovery problem.

Start by identifying the actual control objective. Who must be eligible? Which roles require independent approval? How is access limited, recorded, and revoked? Ask the control owner which decisions require a person, rather than assuming either that every signature is necessary or that automation removes the obligation.

For a defined low-risk role, a possible redesign checks eligibility against current ownership records, grants time-limited access, records its use, and revokes it automatically. Higher-risk privileges retain independent review. Emergency access has an explicit route, monitoring, and follow-up.

Test revocation and failure recovery as carefully as provisioning. A faster grant with unreliable removal is a worse service.

The architecture work is deciding those boundaries across identity, the service, its controls, and its owners. The operational measures then tell you whether the design works in practice.

________

🧭

________

​

Where the human still belongs

Simon Wright challenged Friday’s post: these measures sound like operational improvement.

I agree about the measures. The architectural question is what makes the improvement possible, and what it changes elsewhere.

Someone has to decide which judgments can be made in advance, where exceptions go, and who owns the path when nobody approves each request. Someone also has to check the effect on the operational systems we are changing. A faster front door that moves the waiting or risk downstream hasn’t finished the job.

Eric Arts asked about the role of the human in the loop. That’s the useful question to ask at each stop.

Some people are resolving uncertainty. Others are repeating a settled decision because we haven’t built it into the path yet.

Removing the latter still costs work: reliable inputs, maintained rules, exception handling, evidence, and recovery when the automation fails. Where the decision keeps changing, a well-run human review may be the better investment.

You can remove a routine approval. You still need an owner.

Henry’s road crew remains busy after the traffic starts flowing. The work moves toward maintaining the route, learning from exceptions, and deciding where to build next.

________

🧭

________

​

Take this into Monday’s conversation

Copy this into a team document. Pick 1 path and fill it in with a person who actually uses it.

Path and intended user:

Working result, scope, and service hours:

Sample and evidence: last 10 requests, dates, failures, and open cases.

  1. Required human handoffs: who acts, how often, and what judgment or task each contributes.
  2. Elapsed time: median, slowest completed case, and where waiting accumulates.
  3. Independent completion:__ of __ routine requests. First point where help is needed: ___.
  4. 2 a.m. behavior: what completes, what waits, and how a failed attempt recovers.

Diagnosis: paved road / honest queue / gate wearing a portal / mixed, explain why.

Decision to keep human:

Repeated step to remove or build into the path:

Owner, next change, and review date:

Evidence of improvement: what teams can now finish, time saved, and any change in failures or rework.

Start small. Agree on 1 change, then run the same test again. You don’t need a new platform program to find the first unnecessary stop.

________

🧭

________

​

Community Corner

The comments helped make this week’s argument more useful.

​Henry Luija gave the road-crew analogy a life beyond last week’s newsletter, then joined Wednesday’s conversation about gates that inadvertently block self-service. Henry, seeing you take the idea further was a real pleasure.

Simon Wright pressed on the distinction between operational improvement and architecture, and on the impact on the systems being changed. Both questions belong in the assessment, not just in the reporting afterward. Friday’s discussion​

Eric Arts caught an extra human in Thursday’s illustration: me, in the footer portrait. Fair point.😉 His more important question, the role of the human in the loop, helped sharpen the section above. Thursday’s discussion​

Craig Pinegar and Sunil Mishra picked up the 2 a.m. test. Craig also framed the human dependencies as an adoption measure. That connects the platform’s availability to whether people can actually use it. Thursday’s discussion​

Eptekhar Hasan highlighted the reporting trap: falling ticket counts can hide growing capability. What teams can now do without asking deserves a place beside the volume your team handles. Friday’s discussion​

Thank you all for helping work through the edges of the idea.

________

🧭

________

​

The guardrail download, restored

Last Sunday’s download link was broken. I sent the corrected link separately last week, and here it is for anyone who missed it:

​Download the 5-Field Guardrail sheet​

The Paved Road Test helps you find where a team has to stop. The guardrail sheet helps you write a settled decision into something they can use: the decision, its boundary, the default, an escape hatch, and an expiry trigger.

Use it where the test reveals a repeatable decision. Where the judgment still needs a person, make that route explicit.

I’d love to hear what you find. Hit reply with the path you tested and the first point where someone had to stop and ask.

Water at the tap. Electricity at the outlet. There’s a lot of work behind making something feel ordinary.

That’s work worth showing, even as the noise recedes.

Until next Sunday,

Kevin

________

________

Kevin Donovan​
I write Architects' Hub every Sunday for senior architects and technology leaders.
You're getting this because you subscribed at architects-hub.kit.com.
Know someone who’d find this useful? Forward it to them.

​

Korsvejen 9, Viborg, Viborg 8800
​Unsubscribe · Preferences​

Architects' Hub

A Sunday newsletter for senior and enterprise architects. One decision, one practical model, one tool you can use Monday. Written by Kevin Donovan. Join 2,500+ architects. One email every Sunday.

Read more from Architects' Hub
Three doors slightly ajar in a row. Through them, a telephone, a stack of documents with a magnifying glass, and an arrow curving back on itself. A single coin rests on the floor in front of the doors.

Nov. 22 2026 | Read time: 11 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. The room went quiet Hey , Two Sundays ago, our hypothetical pilot case got its commitment record. Last Sunday it got a walkthrough, and the operating owner signed with 3 tests on record. The AI assistant now drafts delayed-order replies for 3 teams, and the drafts are good. This week someone proposed letting it do more than just draft… In this edition The...

Five simple figures in a ring around a small central box, each holding one object: a magnifying glass, a signpost, a wrench, a mop, a headset. One figure carries a leaning stack of boxes.

Nov 15, 2026 | Read time: 11 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. Tuesday, 10:40 am. Hey ,Last Sunday we left a hypothetical customer-service assistant with a commitment record and unanswered questions about capacity, data and support. This week, move the story forward 3 weeks. The team has agreed to a bounded trial on 1 queue. Agents still review every draft. Wider rollout is waiting. I want to spend this edition with the...

A working terminal with four loose connections labeled Funding, Operations, Data and Authority.

November 8, 2026 | Read time: 9 mins. Kevin Donovan · Architects' HubForwarded this email? Subscribe to Architects’ Hub for the next issue. Hey ,In this edition of Architects' Hub: What the service lead heard when the sponsor said, “We should roll this out.” A commitment record filled in for one example customer-service queue, PLUS a blank copy for yours. How to take a conditional decision back to the sponsor who wants a date. Community Corner: Santiago Jesus Lasso Valenciano on the work that...