What shipped
Newest first. Short entries — what you can now do, and whether it fully works yet. The longer stories about how any of it went live on the blog, and the how-does-this-work pages are in help.
-
The empty cards have pictures now
Three places in the app show you a card explaining something you have not set up yet — garden zones, growing spaces, and the weekly walk. They used to carry a plain gray band where a picture should be. They have real artwork now, and it is doing a job rather than filling a hole: the garden zone card shows several different growing things gathered on one patch of ground, because that is exactly what a garden zone is, and the growing space card shows one empty raised bed with its walls holding the soil up above the grass. Deliberately empty, not lush. A photograph of somebody else's beautiful garden on a card telling you that you have none would be a promise we did not make. We also found two real bugs on that screen while looking at the new pictures: the same button appeared twice, and on a brand-new garden it opened a form with nothing in it. Both fixed.
-
The empty-state pictures lost their backgrounds
The three cards that explain something you have not set up yet had proper little illustrated scenes on them. They now have the objects with the backgrounds cut away, sitting straight on the card. It is a small change and it feels quite different: the thing belongs to the app instead of being a picture of somewhere else pasted into it. The garden zone card changed subject entirely and is much better for it — it is now your garden seen from above, with the house, the fruit trees, the vegetable plot, the raised beds and the pots by the door all on one piece of ground. That is what a garden zone actually is, and looking down at a whole property is the only angle that can show things belonging together.
-
The microphone says who you are talking to
The microphone button on every screen has always handed what you say straight to Rafe, along with the plant you were looking at and any picture you took. Its label said "Say it" — true, and no help at all in working out where it went. It now says "Tell Rafe", and while you are talking the box says "Rafe is listening" rather than just "Listening". Nothing about how it works has changed and it is not a tap slower. Rafe is also a he now, which is a real decision rather than a slip of the keyboard: the app had been calling him three different things in three different places and none of them had been chosen by anyone.
-
The setup card asks, and tidies up after itself
The card that helps you finish setting up used to announce itself and then list everything, done and undone together. Two changes. It now asks — "Let's finish setting up your garden" — and the things you have already done have moved behind a line telling you how many, which you can tap to open. Nothing is deleted and nothing is hidden from you; it is just that on your fiftieth day, a list of things you finished in week one should not be sitting above the jobs you actually have today. The way out changed too. It used to say "Hide this", which describes what happens to our card. It now says "I'll find my own way", which describes what you are choosing.
-
The walk follows your route now
The garden walk used to work one way: photograph up to ten things, then tap which single plant each picture was of. Standing in a bed with one hand free, that is work the app was making you do for it — and most of the time there is no right answer, because a raised bed holds six things and a photo of it is a photo of the bed. So the walk now starts by asking which spots you are going to look at, and asks you to tap them in the order you will actually reach them. Back door first, wherever you finish last. Then it takes you to one spot at a time: it names the spot and what it believes is growing there, you photograph what you see, and you tap Next. Every picture is still looked at, and what we notice is still compared against the last time we looked at that same spot — that comparison was always done per spot underneath, so this matches how it already worked. The walk also now records when you last finished one, which is what the weekly review will compare against. A walk you start and photograph nothing on does not count, on purpose.
-
Your weekly report can make the jobs for you now
The report told you what the walk turned up and then left you to write your own to-do list from it. At the end of it now, it offers: here are the jobs this implies, want them on your list? One job per growing space rather than one per problem, because three things wrong in one bed is still one trip down the garden — and the job carries all the reasons with it. It asks rather than just doing it, and that is on purpose: nobody told us to make those jobs, we looked at your photos and drew a conclusion, and a conclusion is not an instruction. They name the bed rather than a plant, because a problem spotted in a bed does not tell us which plant it belongs to. And on your list they say "From your walk" instead of "You asked", because you did not ask for them — you agreed to them.
-
Undo now tells you when it did not
The "What it did" screen lists anything the assistant wrote or made on its own, and lets you undo it. Undoing a bed it made has always had one rule: if there is still a plant in that bed, the bed stays. That rule is right — the plant might be one you put there yourself, and deleting the bed would take its home with it. What was wrong is what we told you afterwards. The row went gray, crossed itself out and said "removed", over a bed that was still sitting there with your plant in it. Somebody told a thing is gone does not go back and check. Now the bed says so on the row: "Still here — there is a plant in it. Move or remove the plant first." Undoing a plant, or an empty bed, works exactly as before. We found this by tapping the button ourselves rather than by reading the code, which is the only way this kind of thing turns up.
-
What we expected shows up even when you did not photograph it
Every time you write something down, we quietly record what our calendar expected that plant to be doing on that date. Until today you only ever saw that when you had also photographed the plant, because the line was built to compare the two and said nothing at all when it only had one of them. Most entries are a few words with no picture, so on most of your journal the half we always had was being thrown away. Now it says what we expected on its own — "model: flowering" — and it says model on purpose, because nothing was looked at. It is what our calendar thinks, not something we saw, and it will never tell you a plant is on track off a photograph that does not exist. When you did photograph it you get the comparison exactly as before. The main thing this changes is what the app knows when you ask it a question: it can now see what was expected across your whole recent journal instead of the few entries that happened to have a picture attached.
-
When it cannot tell which plant you meant, it asks
Say "I watered the beans" with three lots of beans in your garden and there is no honest way to guess which. The app has always refused to pick one — writing to the wrong plant is worse than not writing at all, because you would have no way to notice. But until today it refused silently: nothing was written and nothing on the screen said so. It now asks, with a button for each plant it might have been, each one named. Tap the one you meant and it goes in. There is no "yes" button anywhere in this — the buttons say the plants' names, because a button reading just "Yes" answers whatever you thought you were asked, and that is not good enough on a screen that writes to your records. And sometimes there are no buttons at all: if two plants share a name and sit in the same bed, nothing we could put on a button would tell them apart, so we say nothing was written and leave it with you rather than offering a choice you cannot really make.
-
When it is not sure, it now tells you how to check
Photograph a spotted tomato leaf and the honest answer is often "early blight or septoria leaf spot" — they look alike, and refusing to pick between them is the right call. But until today that is where it stopped, and being told it is one of two things with no way to tell is a coin flip. The two are treated differently, so the coin decides what you put on food you are going to eat. It now says what to look at: concentric rings like a target inside the spot point to early blight, while tiny gritty black dots in the center of small round spots point to septoria. That is a ten-second check with the plant in front of you, and we were holding the answer the whole time — it is in the extension sources behind every crop page. The field existed and was described as being for when we cannot tell at all, so the app filled it only when it was refusing outright, and left it blank in exactly the case it was written for. It is still left empty when the answer really is certain, because inventing a test to fill a box would be worse than nothing.
-
When it says it wrote it down, it wrote it down
If you had told the assistant it could keep notes without asking, it would answer you and say "already logged" underneath. It was telling the truth about being allowed to and nothing was reaching your journal. The permission was real, the note was not, and the only way to find out was to go looking for an entry that was never there. That is fixed: the note is written before anything on screen says it was, and if the writing fails the screen says nothing rather than saying it worked. Everything the assistant does on its own is still listed under "what it did", and you can still undo any of it. Two things did not change, on purpose. It still only acts on what you actually said — if it thinks you probably meant something, it asks instead of guessing, and asking means nothing has happened yet. And when it cannot tell which plant you mean, it asks which one. Three tomatoes and "I watered it" is a question, not a coin toss, and it will never pick one and write to it.
-
You can tell when it is Rafe
Rafe now signs his replies. Every answer from him starts with a small round mark and his name. Before this, your own messages sat in a purple bubble and his were just text on the page — nothing said who was talking. Nik sent a photo, got a good answer back, and could not tell whether he had been talking to Rafe or to any old chatbot, which is a fair thing to be confused about when nothing on screen says. The mark is a sprig, because Rafe is a gardener. There is deliberately no face — a face makes a promise about how much he understands you, and he will be wrong sometimes. The answer itself still takes the full width of the page rather than being squeezed in beside a picture, because reading it is the point.
-
Your growing spaces can sit in a garden zone
There is now a level between your garden and the beds in it. A garden zone is a part of your garden that holds several growing spaces — your back yard, your front porch, the side of the house. On the garden screen, under Growing spaces, you can group the ones that sit together and give the place a name. You do not need to: every growing space works on its own exactly as before, and most gardens will never group anything. It is there for the moment you find yourself saying "the whole front porch" and meaning four pots at once. Taking a grouping apart puts the growing spaces back on the plain list and touches nothing growing in them — there is deliberately no way to delete a bed from that screen. One limit worth knowing: garden zones do not nest. A back yard and a back yard orchard are two separate places, and a tree in the orchard is not also in the back yard. We chose one level so that every total stays simple and cannot count the same tree twice.
-
Your plants say who added them
A plant Rafe created now says so on its own page: "Rafe added this", followed by how it was allowed — you said yes to this one, or you let him write things down, or you let him do everything. Months after the fact, looking at a plant and wondering whether you put it in or Rafe did is a real question, and the answer used to be nowhere at all. We were recording it and never showing it. There is deliberately no undo button on that line: the page already has "It is done" for a plant that finished and "Delete this plant" for something added by mistake, and those are genuinely different things. One generic undo would have to guess which you meant.
-
Your week, in a few cards
Finishing a garden walk now gives you a report: a short set of cards you swipe through, rather than a list. What is wrong, anything that is not where we would expect it to be, what we could not tell about, and what is coming. That third card is the one worth knowing about, because most gardens are mostly on it. We can only say a plant is on track when what you photographed and what our calendar expected happen to use the same word for it, and most of the time they do not. So we say we could not tell rather than saying nothing at all — silence would read as "fine", and showing you the gap is more use than inventing reassurance about it. A quiet week still gets a report, on purpose. Knowing we looked and found nothing is a different thing from us having said nothing, and it is most of what a garden year actually is.
-
A hot spell now covers the days it lasts
When a run of hot days is coming, the weather strip marked it on the day it started and nowhere else. A four-day spell looked like a Thursday problem. The orange band now runs across every day the spell covers, so you can see how long it lasts without tapping anything. What it does not do is repeat itself: the word HEAT and the number of plants are still said once, on the day it starts, because a four-day spell is one thing to plan around rather than four separate problems. That was a deliberate decision when the warnings were built and it has not changed — only how wide the mark is drawn. The same goes for what a screen reader hears: one day announces the count, and the rest say only that they are inside the spell. Frost is different and stays different. A run of cold nights is a run of nights you go out and cover, so each one is still marked on its own rather than joined into a single stretch.
-
A plant names itself
Adding a plant no longer asks you to think of a name for it. A cherry tomato is called Cherry Tomato, and the next one is Cherry Tomato 2. Tap the name on the plant to change it to whatever you actually call it — "the one by the shed" works fine, and that is the name the app uses from then on. The numbering counts across your whole garden rather than per bed, so two plants never end up sharing a name and leaving you to work out which one a note was about.
-
A tree we cannot age is not told its flowers will drop
If you sowed a plant, its planting date is its whole life, so we know what it is doing. If you bought a tree, the planting date is the day it arrived — it may already have been three years old. We were reading a years-long clock off that date, deciding a young tree was in flower, and warning that a hot spell would cost it fruit it was never going to set this year. One gardener asked, reasonably, whether his orange was even supposed to be flowering. It was not. Now, when a plant lives for years and we have no way to tell whether it is old enough to bear, we say nothing about its flowers. The heat warning still appears and still gives the two things that help — water in the morning, shade in the afternoon — without a claim we cannot stand behind. Nothing changed for annuals: tomatoes, beans and squash are unaffected, because for those the sowing date really is the clock. And this is a floor rather than a ceiling. As we source how many years each fruit tree takes to crop, the same warning goes from silence to a real answer, including telling you when a tree is old enough and the heat genuinely does matter.
-
Ask for a plant by its variety
You can now name a plant by what it is rather than only by what it is called. Ask about the Sungold and it finds your Sungold, even when that plant is listed as "Cherry Tomato 2" and nothing on it says Sungold. A variety you typed in yourself works the same way — your own words are the only record of it, so they are searched too. A name you gave a plant still wins over a variety, and a variety beats the crop, so asking about the Sungold does not come back with every tomato you own. This matters more now that plants name themselves: when three are called Cherry Tomato, Cherry Tomato 2 and Cherry Tomato 3, the variety is often the natural way to say which one you mean.
-
Asked where the job is, not on the way in
The day band can say more if it knows roughly when you are usually home — it can point at hours you could actually use, instead of hours you are never there for. So it asks. But only at the moment that knowing would help: when there is something on the band worth placing and nothing to place it against. Not on a quiet day, not on your first run, and not sitting at the top of the screen every morning waiting to be noticed. It asks once. Dismiss it and it does not come back, and dismissing counts however you close it — the button, the background, or the back gesture. A question that only takes no for an answer if you press exactly the right thing is not really asking. The times themselves open right there on Today rather than sending you off to Settings to find the right section, and it is the same control that has always been on your profile page, so answering in either place is answering once. None of it is required. Skip it and the band works exactly as it did.
-
Choose how much Rafe does on his own
Until today every gardener was on the same setting and nobody could change it. Now there are three, in Profile: ask me before writing anything down, write notes on his own but ask before making things, or do everything on his own. You start on the middle one. The prompt that asks about a new bed also has a shortcut — "Make them, and stop asking me" — which makes the bed and moves you to the last setting; Profile is where you put it back. Whichever you pick, you can always see what he did and undo any of it, which is true on every setting rather than a feature of one.
-
Faint text on the plant pages
Some small print on the plant pages was too light to read comfortably against its background — the "Do now" and "Look for" labels, the note about how often to water, the region rows under the fall sowing dates, and the little "est" tag that marks a figure as our estimate rather than a published one. All of them are darker now. Two of those were the same mistake made twice: the text was already set in the quiet gray-brown we use for secondary information, and then faded again on top of that, which took it past quiet and into hard to read. The second fade is gone rather than reduced. The status tags on the help index — Live, Partly live, Not built yet — had no color of their own, so they were picking up the color we use for links, which was both wrong and too pale on the green tag. They now have their own dark ink and are told apart by their background, which is what those colors were for.
-
Four sentences that said one thing
Tap a hot day and the screen told you about each plant at risk on its own line. When four plants were in the same trouble that meant the same twenty-two words four times over, differing only in the plant’s name and a number — sixty-seven words where twenty-six would do, on a panel that was already cutting off its own last paragraph. Plants in the same trouble now share one line: the plants named, the range of how many hot days each has had, and the reason once. What it will not do is join up plants whose trouble is different. A crop running to seed and a crop dropping its flowers are two different things happening for two different reasons, and squeezing them into one sentence would be wrong about both — so those keep their own lines, and so does anything where the wording would have to be invented to fit.
-
He asks before he makes something
Ask the assistant for a new bed and he now stops and asks you, with the bed he has in mind and its size, and two buttons. Writing a note down and creating something are different: he writes the note as he answers, and he asks before he makes. A reply can do both at once, and when it does, the part already written sits above the answer as a finished sentence and the part waiting on you sits below it. The buttons say what they act on — "Make the bed and add the plant", not "Yes" — because on a reply with two halves a bare yes or no could be read either way, and saying no to a bed has never touched anything already written down. A plant and the bed it goes in are one question: one tap makes both, and the plant goes into the bed it just made.
-
It says when it does not know how a plant is planted
Moving a pot into the shade beats shading it where it stands, so a hot-spell warning tells you to move anything it knows is portable. If a plant has never been told how it is planted, it used to land in the same bucket as something rooted in the ground — the advice quietly dropped to "shade in the afternoon helps" and nothing said why. Now the card says what it does not know, and asks: in a pot, or in the ground? Answer it on the plant and the warning changes straight away. We deliberately do not guess this from the bed. A bed can be marked as a container and still hold a plant growing in the ground — a front porch is a place, not a pot — and guessing from it would mean telling somebody to pick up a tree. The plant is the thing that knows how it is planted, so the plant is what we ask. Nothing changes for plants you have already placed, and a garden where everything is placed sees no extra line at all.
-
It writes the note down while you are talking
Say "I watered the beans this morning" and the assistant writes that in your journal as it answers, instead of making you log it again afterwards. The thing we said on 26 August was not built yet is built now. It tells you in the same reply: a line above the answer reading "Already logged — the beans: watered this morning." You are never told about it afterwards, or only in your journal, because a turn that both writes something and asks you about something else must not let the writing go by unmentioned. Making a bed or adding a plant still asks first — writing a note down is not the same as creating something, so it does the first on its own and asks before the second. And the entry still says the assistant wrote it, so months later you can tell that apart from what you typed.
-
One disease was filed under two names, so half the answer went missing
Blackleg is a rot that girdles the stem of cabbage-family plants. In our plant records it had been written down two different ways — one spelling on kohlrabi, another on brussels sprouts — and nothing connected them. So if you asked what else gets blackleg, you were told about one of those plants and never the other. Both spellings meant the same disease. They are now one entry, and it correctly shows up on both plants. Every other disease those two share — leaf spot, black rot, clubroot — already matched, which is how we knew this was a slip rather than a difference somebody meant. Worth saying plainly: nothing you had written down was affected, and no plant of yours changed. This was our reference data being wrong about itself.
-
One thing does not need a headline
When you had exactly one job today, Today showed a large orange bar announcing "1 thing today", directly above the one row it was announcing. The bar is a summary — a count, and a button to start everything at once. With one job the count is something you can see by looking at the row, and starting everything at once is just starting the thing, which that row already offers. It was a loud second door into a room you were already looking at. So at one job the bar is gone and the job simply sits under the "Garden tasks" heading. At two or more it comes back, because then it is doing real work. This came out of a screenshot rather than a test. Every block on that screen had earned its place on its own merits, and nobody had stood back and looked at the sizes together — 540 pixels of context above a 67 pixel announcement of a 143 pixel job. Individually reasonable, and as a whole it told you the context was the point and the job was a footnote, which is the opposite of what that screen is for.
-
Our own warning argued with our own all-clear
On one screen, the day band said three named plants were over their limit from noon. Directly underneath, the week's summary said nothing we can check is at risk. Both sentences were honest about their own rule, and only one rule was right. The temperature we hold for each crop is not a figure for one hot hour. The sources it comes from describe harm building over consecutive hot days — "when day temperatures exceed" — and our own notes on that field warn, in so many words, that using it as an instant trigger will fire on an ordinary summer afternoon. That is exactly what the band was doing. The week's summary was applying the rule properly and was right to stay quiet. So the band no longer decides whether anything is at risk. It asks the part of the app that already makes that call, and then does the one job it can do honestly: showing which hours of the day the trouble falls in, for a plant already judged to be in it. If nothing is at risk, the band marks nothing, however hot the afternoon gets. Two parts of an app disagreeing in public is bad. The part that was wrong being the one that shouted is worse, and that is the half we fixed.
-
The checklist gets out of the way once you have started
The setup list on Today had two shapes. A full list, for somebody who has just arrived and needs to be walked through it. And a single line you can tap, for somebody who has nearly finished. That was the right idea, and it switched at the wrong moment: the line only appeared when exactly one thing was left. So a garden that was three-quarters set up still got the whole list every morning, including the three items already ticked, with a progress bar over the top. It now switches the moment you finish your first item, which is the point where you stop being new. Nothing is hidden — the line still says how many are left and takes you straight to the next one, and a brand new garden still gets the full list, because that is the one time it is the most useful thing on the screen. What this actually buys you is room. On a phone, the day's jobs used to start below the bottom of the screen, so the one thing the page exists to tell you needed a scroll to reach. Now they are visible when it opens.