Restech 2.0: The Co-Operator

I spent the last few months implementing every new AI tool, feature and agent that shipped, and something worth trying shipped every week. If a tool couldn't do something yet, I waited a week, then it could. It was exciting, because piece by piece, these tools were solving problems I'd spent six years watching operators face at Rezdy, Checkfront, and Regiondo.

And that became the trap. Running in every direction takes you nowhere. We're all so busy trying the new tools that nobody stops to ask the only question that matters: where is all this actually supposed to take us?

So I stopped to think about it properly: where we were, where we are now, and where all of this is actually going.

Who makes the decisions defines the era:

Restech 2.0 is our best educated guess at where this industry is headed. It comes from three places:

We took our time with it, because our business is on the line: we had to work out where the industry will be, and how to best serve operators then. At the end of the day this is just our take, and you're free to disagree.

Hopefully the rest of this piece shows you how we reached the idea. Fair warning: a lot of time, thought and reasoning went into this, and the article is just as long, about a 25-minute read. The map below lets you jump anywhere.

What Restech 2.0 and "co-operator" actually mean

Here's the full definition, and then I'll unpack the words that matter.

You've heard the word "agents" everywhere this year. This is what one looks like working a tour business. "Standing" matters: you don't grant permission task by task. You brief it once, and the authority holds until you change the rules.

In practice it's a three-step handover:

1. You set the rules. Prices, policies, limits. The business stays yours.
2. It acts inside them. Decides, executes, replies, without waiting to be asked.
3. It escalates the rest. Money, promises in your name, cancellations, safety, exceptions to your own rules. The owner's calls, always.

It's the difference between a tool and a staff member. A tool waits for your hands. A staff member you brief once, and they get on with it, coming to you only when something needs the boss.

That's why co-operator is the right word, and I did try the others.

Co-operator keeps the operator in the word, because you stay in the arrangement. A colleague made of software. I'll use that word from here.

In fact, read the definition from the other side and it collapses to something even shorter. Restech 2.0 isn't software that got better. It's the era your business gets a co-operator. 1.0 is a tool. 2.0 is a co-operator.

Or in eight words: 1.0 has your instructions. 2.0 has your agency.

Restech 2.0 is not a product. It's a system: the tools and software you already use, orchestrated by an AI like Claude, and later I'll show you from which pieces. What this piece offers is the picture of where I think our industry's technology should get to in the next two or three years: using everything AI can genuinely do, while keeping the human element exactly where it belongs. We work in the most human industry there is. Nobody books a tour to meet software.

Sometimes it's easier to see the change side by side, so here's the same shift as a set of pairs:

The current generation follows instructions. The next generation understand it.

And those six years at Rezdy, Checkfront, and Regiondo matter to how this piece is written, because I want to be fair to Restech 1.0: it's genuinely good, and I helped build and sell it. A proper 1.0 system holds your calendar, takes payment at midnight, stops the double-booking, sends the reminder that halves your no-shows. I watched it give operators their evenings back.

But I also watched, year after year, the problems 1.0 couldn't touch no matter how many features we shipped. A guide calling in sick. One late pickup delaying the entire tour. Every action a 1.0 system takes can be traced back to a decision a human made: in advance, as a setting, or in the moment, as a click. It moves only when asked, and whatever the command didn't cover is still your problem, and the system doesn't know your problem exists. Those are exactly the problems the new tools can finally reach. Not as a better version of what we built. As a change in kind: a fundamental shift in how you and your restech work with each other.

Here's what that looks like, on a morning you already know.

A sick guide at 6:40am, on today's software

A voice note lands at 6:40am. One of your guides is sick. He's sorry. He's not coming in.

You know this morning. Here's what it looks like with the software you run today.

Here's the part worth noticing. Your booking system knows everything about this morning except that there's a problem. It knows the sessions, the guests, the payments, the policies. It watched the whole thing and said nothing, because nobody clicked anything. In today's software, information you didn't ask for is information you don't get.

Don't ask, don't find out.

Four behaviours that separate 1.0 from 2.0

That morning didn't fail in one way. It failed in four. The system never spoke first, so the problem was yours to find. You were the only thing connecting cover to messages to replies to refunds, so the chain lived in your head. The cross-checking at 9pm was you doing the summarising. And the one guest who needed the boss got sat on, because escalation had no shape. Four failures, and the definition you read earlier promises exactly their four inverses. They're not a feature list I invented.

Handling any problem, whether the handler is your best staff member or a piece of software, has four parts: noticing it, understanding it, working on it until it's done, and knowing which calls aren't yours to make. That's initiating, condensing, chaining, escalating. Do all four and you're handling problems. Miss any one and you're helping with tasks.

It initiates. Nobody asked it anything, and it speaks. "Thursday has a conflict you haven't seen yet." In 1.0, problems only show up when a guest is standing at the jetty. This is don't-ask-don't-find-out reversed: information you didn't ask for finally reaches you while there's still time to act on it.

It condenses. Many sources become one summary. Forty messages become "three need you." That requires understanding what matters, which takes judgement, not just a search. This is the 9pm problem reversed: the cross-checking you were doing at the kitchen table is exactly the work being condensed.

It chains. It follows one event through every consequence, step by step, until the problem is closed, not just the task. Each output becomes the next input, and the chain runs across hours and days without anyone asking again. This is the five-tabs problem reversed: in the 1.0 morning, you were the chain, the only thing connecting cover to messages to replies to refunds.

It escalates. The protected calls always come back to you, presented as one plan you can approve in a glance, not forty notifications you have to assemble yourself. This is the sat-on exception reversed: the request you couldn't find a minute to think about all day comes to you with a suggested answer attached, not as another open tab.

Restech 2.0 is all four together. They lean on each other: a system that chains but can't condense produces work you can't audit. One that initiates but doesn't escalate is a liability with a voice. One that escalates without condensing is just forty notifications with better branding. The behaviours only earn trust as a set, which is exactly why no single add-on feature can fake them.

Four claims are easy to write down. So here's that same sick-guide morning again, run end to end. And after it, a slower problem, because a test that only works in a crisis isn't much of a test.

The same morning, with a co-operator

Same voice note. 6:40am. Guide is sick. This time a co-operator has the morning.

Before you've finished your coffee, it has worked out the damage and tried cover first, because it knows that finding cover is a smaller disruption than rebooking fifteen people. That ordering is a judgement call, and it made it without being told this morning. The chain has started.

And here's the part that makes it trustworthy rather than terrifying: everything gets condensed to one screen before anything goes out. You read it in a glance and say go. That screen looks something like this:

From there the chain works the replies all day while you run the sessions that still exist. The one guest asking for something outside your policy escalates back to you, because exceptions to your own rules are always yours. The chain closes when all fifteen guests are resolved, and you're told so.

Notice what changed in the sick-guide morning. It's not that any single step was impossible before. It's that nobody was holding the problem open. In the 1.0 morning, you were the thing connecting every step, across five tabs and three days of guest replies.

1.0 closes when the command is done. 2.0 closes when the problem is solved.

One honest note before we go on, so this scene doesn't oversell: parts of that morning can be built today, the reading, the drafting, the summary you approve. The full chain, running on its own across a whole day, is the two-or-three-year picture. That gap is exactly the road this piece is named after.

A real operator: Satu's casual-driver problem

That morning is mine, invented to make the point cleanly. Here's a story that isn't.

Satu runs Let's Go Discovery, jeep tours in the Cameron Highlands, and his drivers are casual workers. That single fact shapes his entire booking flow: he can't auto-confirm anything, because a booking is only real once a driver can take it.

And the confirmations were always inside his rules, because the rule is his: a booking is real when a driver says yes.

Notice there was no rostering software in that story. The co-operator asked the drivers themselves, on WhatsApp, and their yes was the data. Instruments don't have to be products. Sometimes the source of truth is a human who knows whether he's free on Saturday.

Of all the stories in this piece, this one is the closest to being buildable, which is exactly why it carries a real name. The road isn't all two or three years away. Some of it is next season.

Spotting a booking decline months before it costs you

When a crisis hits, you know straight away. But some problems build up quietly, and you don't see them until it's too late.

A trekking operator now. Her trips get booked months ahead. For weeks something has felt slightly off, but no screen was ever going to tell her. A problem like this is a leak, not a fire. In 1.0, you only find a leak when it shows up in the accounts. In 2.0, it proactively warns you.

She reads it over coffee and picks one. A minute, not an evening, and a whole season still on the clock.

Then the behaviour that separates the eras: it doesn't leave her a chart and move on. It carries the choice out, watches the pacing, and comes back either way: recovered, or not, and if not, with the next option already prepared. The pricing call, being money, was always hers.

Same four behaviours, no drama. It noticed without being asked. It condensed a season of numbers into one sentence. It followed the fix through to an answer. And the money stayed with the owner. The sick guide tested whether it can handle a bad morning. This tests something different: whether it's watching at all when nothing is on fire.

What each era looks like on a screen

Each era has a look, and the look tells you what's underneath.

Pre-Restech looks like a notebook. A hardcover by the phone, a whiteboard on the wall, and your memory holding the rest together. The interface is your handwriting.

Restech 1.0 looks like a dashboard. The calendar grid, this week's revenue, and the best example of all: the daily manifest. The manifest is a genuinely great screen, and it's worth understanding why. It works because every day asks the exact same question: who's coming, what time, how many, paid or not. Someone could design that screen once, months in advance, and it's right every single morning.

Which is also its limit. A dashboard is a bet, made months early, on what you'll need to see. The manifest wins that bet because the question never changes. The sick-guide morning loses it instantly: one event turns into fifteen different guest situations, and no pre-built screen exists for that shape of problem. That's why, the moment a 1.0 day doesn't go to plan, you leave the dashboard and end up living in browser tabs and WhatsApp threads. The software's screens simply weren't designed for the problem you're having.

Restech 2.0 mostly looks like a conversation. A chat thread. In a screenshot, honestly not much more impressive than WhatsApp. That's one of the strange things about this era: it looks less impressive in a demo than the last one. A 1.0 dashboard looks great in a screenshot. A 2.0 interface looks like texting a very capable staff member.

Until you need to see something. Then the right view gets built for that exact moment: the whole sick-guide plan laid out on one screen, a chart when a chart helps, three words when three words are enough. And then it's thrown away, because tomorrow's problem will need a different view. This isn't a distant future, either: glasses that put an AI's answers in front of your eyes are already on sale today, and the screens they show are exactly this kind, built for the moment, gone the next.

That's the shift in one line: in 1.0, we designed dashboards ahead of time to show the same things every day. In 2.0, the view is generated on demand to fit the moment.

Let me be honest about the limits, because this piece only works if it's honest: today's AI is good at choosing the right view and getting better, but it is not solved. It will still sometimes hand you a spreadsheet when you needed a sentence. That's part of what has to mature before the two-or-three-year picture is real.

Four tests to know if a system is really Restech 2.0

New language spreads faster than new software, so over the next couple of years you'll see plenty of products described in words like the ones in this piece. Some will be the real thing. A definition is only useful if you can check it, so here are four questions that tell you whether a system, any system, ours included, is really Restech 2.0. Each one is checkable in a demo.

And one caveat that matters more than any feature list: proactive alone is not 2.0. Push notifications have been "proactive" since 2010. An alert on a threshold is just your own instructions running on a delay. An app that pings you isn't smart. It's just following an alarm someone set. The real thing knows what matters from understanding your business, not from a rule somebody configured in a settings screen.

The same test catches workflow automation wearing the label. An automation is your instructions running while you're not in the room: genuinely useful, and plenty of operators run good ones. But every branch of it was drawn by a human in advance, and the moment reality steps off the flowchart, the automation stops and the problem is yours again. The dividing question is simple: when something happens that nobody wrote a rule for, does the system work the problem, or wait for you? An automation follows a path someone drew. The real thing draws the path.

And to show the tests doing real work, run them on a real example: the AI inbox assistant, probably the most useful new tool of the last two years. It reads your overnight messages, sorts them, drafts replies for you to approve. Genuinely worth having. Now score it. It condenses, beautifully. But you speak first, because it runs when you open it. It closes commands rather than problems, because once the drafts go out, whatever happens next is yours again. And it escalates nothing, because it holds nothing back to escalate. One behaviour out of four: a very good 1.0 tool wearing this era's clothes. That's not an insult. It's the honest score, and it's the score to give every product that claims the label.

When a 1.0 product bolts on a chatbot and claims the era, the tests catch it. The chatbot speaks second, executes one step, hands the problem back to you, and its "proactivity" is configured alarms. No single feature fakes all four behaviours at once.

What a co-operator is made of: a brain and instruments

Look at what the sick-guide morning actually required. Something knew the guide was sick. Something knew which other guides were qualified and free. Something knew which vehicle passed this morning's inspection. Something knew every booking, every deposit, that one promo. Something knew which refunds actually went out and had to land in the books. Something drafted in your voice. No booking system holds all of that, and no booking system should try, because the tools that do each of those jobs well already exist.

So the working shape of 2.0 is a brain, and instruments. The brain is a general AI, the same assistants millions of people already use, a Claude or a ChatGPT, connected to your tools and briefed on your rules. The instruments are the products that each hold one part of your operation, and it's worth being concrete about what each one actually is and why the brain needs it.

Your rules

Your restech The hands

Availability, guest comms, payments

KongRezdyCheckfrontBokun

A general AI: the brain.

Briefed on your rules.

ClaudeChatGPT

Rostering

Who's qualified and free

DeputyWhen I Work

Safety and assets

What's inspected and safe

SafetyCulture

Accounting

What's owed and reconciled

XeroQuickBooks

That last part is the restech's job in the new era, and it's a big one. When the roster says a guide is out, or the asset tool fails a vehicle, the restech is the instrument that turns that fact into reality your guests can see: capacity adjusted, sessions closed, the right guests contacted, refunds and reschedules moving under your policies. The brain conducts. The restech is the hands for everything booking-shaped.

So why shouldn't one product just own all of it? Because each of those jobs is deep enough to be a whole company, and in every case it already is one. Rostering done properly is pay rules, fatigue limits, qualifications and shift swaps. Safety done properly is audit trails and compliance that stand up when an inspector asks. Accounting done properly is tax. Every time a booking system has bolted on a shallow version of somebody else's specialty, operators have ended up running the real tool anyway, plus the add-on. You already run a stack of specialists, and that was never the problem. The problem was what connected them. In 1.0, the integration layer was you: you read the roster, remembered the failed inspection, retyped the numbers, at 9pm, at the kitchen table. What changes now isn't that the tools merge. It's that the brain can talk to each one in its own language, so they never have to. The tools stay specialists. You stop being the messenger between them.

The fair question is whether the specialist tools will play along, since being conducted might sound like giving something up. I think the incentive runs the other way. Once the brain routes the work, being a tool it can talk to becomes the new minimum, the way having an API was, the way having a website was before that. The instruments that speak get pulled into every stack a brain assembles. The ones that stay silent fall out of it. Nobody has to cooperate out of goodwill. They'll cooperate because that's where their customers will be.

That also changes what "your restech" means in the definition above. In 1.0, your restech was one product you logged into. In 2.0, it's the system: the brain plus the instruments it conducts, with your rules standing over all of it. That whole arrangement, working as one thing, is your co-operator. Which means the most important sentence in this section is this one: nobody sells you a co-operator. You assemble one, out of pieces that each do their job well and know how to talk to each other.

And this is my favourite thing about the whole picture: it makes the era impossible to own. Restech 2.0 is a destination the whole industry builds toward together, each product doing its part well and learning to speak to the rest. That includes us. Kong's job in that picture isn't to be the brain. It's to be the best pair of hands the brain has ever had.

The technology arriving now that makes Restech 2.0 possible

Underneath everything in this piece sits two shifts.

Because both sides are freed at once, this era gets an agent rather than just a better app. Delegation is a loop: you have to be able to hand something over mid-task, and see what came back at a glance. Free input with fixed screens still drags you to a desk. Flexible screens you can only command by typing still need your hands. Only both together make a staff member rather than a tool.

Restech 2.0 will feel like literal magic

Imagine you're wearing a pair of AI glasses.

In 1.0, the screen was the revolution. In 2.0, the screen is the limit.

The five things Restech 2.0 cannot replace

Five items, one rule underneath: anything that creates an obligation your business can't easily take back belongs to the owner.

Watch the rule work on a genuinely unclear case. A guest asks to move their booking forward a week, onto a date your early-bird promo happens to make cheaper. Reschedule, or exception? Split it with the rule.

One request, split correctly down the middle, and nobody needed a policy meeting.

None of this should feel new, because it's exactly how you already run your team. When you hire someone good, they decide freely inside their jobscope and bring you everything outside it. And when they bring it, they don't arrive empty-handed. They come with a suggestion: this guest wants a refund outside policy, she's a repeat customer, I'd offer a reschedule with the fee waived, okay? You make the final call. They carry it out. You didn't do the work, but the decision was never anyone's but yours.

That's exactly the shape of a 2.0 escalation. It arrives as a prepared plan you can approve in a glance, not a problem dumped back on your desk. The software has already done the thinking a good staff member would do. What it hasn't done, and never should, is make the owner's call.

That's not a feature preference. It's the law of the era. A system that never asks isn't further down this road. It's off it. This era isn't named after autonomy. It's named after trust.

The point of all this was never to remove you from your business. It's to remove you from the parts of your business that were only ever admin, so the judgement calls actually get your full attention instead of arriving at 9pm, at the kitchen table, underneath forty unread messages.

When the co-operator gets it wrong

Because it will. Software has bugs, judgement calls miss, and even the staff member you'd trust with the keys occasionally books the wrong van. The question that matters isn't whether a co-operator makes mistakes. It's whether its mistakes are built to show up.

That's what the behaviours and the tests were quietly pointing at all along. Anything with real consequences shows up before it goes out. Everything it did sits in one glance afterwards. Anything touching money, promises, safety or your own rules waited for you anyway. So a mistake inside your rules gets caught the way you'd catch a staff member's: visible in the plan, correctable in the rules that produced it. And a system that can make invisible mistakes has failed the trust test, which means by this piece's definition it isn't 2.0, whatever the label says.

One thing delegation never moves, though: the responsibility. It's still your business and they're still your guests, which is exactly why the escalation list exists and why the era is designed around you seeing everything.

The parts of Restech 2.0 that exist today

Honestly: partway, and unevenly.

The pieces that read, sort, summarise and draft are genuinely useful today. An operator can set up an agent this afternoon that reads the overnight inbox and hands them a morning summary with replies drafted. The pieces that predict and the pieces that move money and bookings are further out, and some of what's being sold right now under an AI label is not worth your time.

And since I've asked you to hold me to a timeline, here's what holding me to it should look like.

If those three aren't real within two or three years, this piece was wrong, and you're allowed to say so.

That gap between the hype and the useful is exactly why I'm not just publishing this piece and walking away. Every week, my team and I test the new pieces on our own business, and we publish the verdicts, including the failures, in a short newsletter. It's named after the destination. When we say a piece isn't ready, we say what would change our minds.

And yes, Kong is building its part of the picture above: the restech's role in that system, the hands. Of course we are. We took the time to work out where this industry's technology should go, and if we then didn't build toward it, you'd be right to ignore everything above, because we wouldn't believe it ourselves. But let me be equally plain about the other half: by this piece's own definitions, Kong today is a 1.0 system, and as far as I can see, nobody is at 2.0 yet, because 2.0 isn't a thing any one of us can ship alone. That's why it's a road. You'll hear about our part when it's real, and not before. Restech 2.0 doesn't require Kong, and the day anyone tells you their brand is the definition of the era, apply the four tests, including to us.

Blake Ng is the co-founder of Kong, a booking platform for tour and activity operators. Before Kong, Blake spent six years at Rezdy, Checkfront, and Regiondo in product marketing. He's based in Kuala Lumpur and personally answers operator questions on WhatsApp, because a question from an operator is exactly the kind of thing that should always reach a human: +60 12-429 8159.