Tahle kniha na mém seznamu k přečtení byla dlouho, bohužel už ani nevím, kdo ji tehdy doporučil. Hezky odsýpá, kapitolu na stránku či dvě, každá popisuje jak to dělali oni, když zakládali BrewDog.
A tady pár odrážek, které mě z knížky zaujaly:
pravděpodobnost, že začínající firma přežije, je malá, a pokud má jednoho zakladatel, je ještě menší;
nehledejte díru na trhu, vytvořte si vlastní trh;
neztrácejte čas s podnikatelskými plány, stejně se dynamicky mění;
nesnažte se prodávat, dělejte skvělý produkt a lidé se o něm dozví (no, tohle mi přijde jako hodně riskantní);
je dobré mít nepřátele, dokazuje to, že jste se v životě někdy za něco postavili (Winston Churchill);
starejte se o čísla a rozumějte financím, králem podnikání je hotovost;
nenabízejte jenom jednu cenu, udávejte cenové rozpětí (napříč různými balíčky) a vždy přidejte nějakou bláznivou volbu;
rozdávejte – BrewDog zveřejňuje přesné recepty a prodává kutilské soupravy, se kterými si zákazníci mohou uvařit piva doma;
firemní kultura tvoří třetinu úplně všeho, je to to, co firma dělá, když si myslí, že se nikdo nedívá;
lidé potřebují vědět, proč firma dělá to, co dělá, musí znát důvody rozhodnutí. Teprve potom se ztotožní s cíli a budou neúnavně pracovat – každý týden rozesíláme lidem email, aby všichni věděli, co se děje;
pohovory stojí za houby, nabídněte uchazečům něco k pití, může to prolomit ledy (vodu ani čaj zjevně nemyslí);
nečekejte, až někdo projeví o práci zájem, ty, které chcete ve firmě mít, musíte objevit, oslovit a přetáhnout;
slavte. Slavte pořád, každý úspěch!
firmy žijí a zanikají podle toho, jak dobře se rozhodují při přijímání zaměstnanců. Když přijmete hodně pitomců, stanete se jedním z nich;
pokud jste vždy mnoha způsoby k zastižení, je víc než pravděpodobné, že se nakonec budete jen točit v kruhu;
čím víc reakitvní práce (vyřizování emailů) vykonáváte, tím víc si jí vytváříte;
firma BrewDog má pro své ředitele pravidlo padesát na padesát – polovinu pracovní doby smějí strávit každodenním provozem, druhou musí věnovat zlepšování, růstu a rozvoji firmy;
důležité je odpuštění, ne svolení;
pět! Zapište si svých pět největších problémů, sedněte si s týmem a vyřešte je. Potom dalších pět;
chcete -li svým burácením vyvolat odezvu, nekřičte moc času – před zásadními tiskovými oznámeními si dejte klidně pár měsíců času;
nenechte se řídit výborem, brainstorming je dobrý tak pro vymaštěnce, spolupráce je pro lidi bez nápadů, kteří se zoufale snaží v podstatě z ničeho vykouzlit přihlouplé nesmysly;
nejste-li nadšení hned od začátku, nikdy to dál nedotáhnete (Steve Jobs).
Dreamforce is over, long live Dreamforce! Every single year the conference is different to me and this year was no exception, if for nothing else then for these two reasons – such a bright idea to do a mini vacation at Yosemite before the event and to record daily podcast during the event. Which forced me to take notes of those a few things I actually visited. It wasn’t the best thing I ever recorded, what would you expect from a piece which is being recorded at 11pm after drinks and just shortly before landing into a bed.
Yosemite
It was Dreamforce #6 or #7 but I never though about extending the event much. It was always a day or two in San Francisco, maybe Alcatraz visit or just walk in that huge park, but nothing more. This year I found out there are national parks just short (4h) ride away and the plan was born.
Sadly I wasn’t alone as 4 millions (or more?) people visit the park per year, so all the camping sites have been reseved as well as most of the hotels. But just outside of the park I found one of the very last rooms and booked it.
What can I say – the experience was totally worth it, the park is huge and even we lived just outside of the park we still drove about an hour or longer every day to get to the main spots.
30 000 steps per day was the norm, 3 days, 3 different locations and experiences. From waterfalls which didn’t have much water, compared to their spring versions, but were still awesome; to above the valley views where one cannot believe that the day before we’ve been 600m lower climbing up; to a completely abandoned valley where we’ve met just two hikers on our way back.
Experience to repeat, next year some different park. Yellowstone, Grand Canyon or something else?
Dreamforce Day #0 aka MVP day
While most people just arrive on this day, it is a long tradition that Salesforce MVPs have their own un-conference, where we meet each other (!) and selected product teams from Salesforce, giving first (or last before go-live?) feedback and being pretty direct.
We typically have several session to choose from and this year I decided to give (again) chance to Data Cloud, Agentforce and Revenue Cloud as I somehow didn’t feel I already have enough of those.
We spoke about the need for clean data, which is a topic Salesforce takes for granted and don’t really amplify it, customers are then surprised their results are slightly different to what they saw in demos. I’ve been also surprised by the huge ratio of zero copy approach (more than 30 % of data are served this way).
The need for additional and deeper enablement of partners and in general and the trouble of staying on top of the things with monthly or even shorten release cycle. And the final thing was Clean Room for possible partner collaboration on data which looks super promising but I wonder how many of my customers will have a usecase for that.
Speaking about Agentforce – 95% of POC will never make it to production, which might lead us to a question whether we should really be lean or less lean if there would be higher ROI.
New AgentBuilder with their own AgentScript as a configuration language which should be easier to use/debug as it won’t result in tons of always/never/before statements which contradict each other. Also new Prompt Builder with conditional control.
Revenue Cloud new features around contract management where I can upload the contract and then ask AI questions around it and what’s covered and what not, soon to be live Compliance Agent, agents which checks why some product is not available for users to choose and fix the potentional configuration issue. Plus it is API first and build on core.
The day is a lot about meeting others as well and it is so great to see all of them in person, at least once a year. For me also the main reason why I finally decided to go to Dreamforce, I’ve been really scared with the situation there, the stricter border control and also the government shutdown – at the end no problem at all.
One main experience of the day – rock, paper, scissors in groups – what an unexpected fun.
But besides meeting MVPs I also got the chance to meet CTAs and if I’m not mistaken this was the first official gathering since I became one of them. No picture probably survived, but the food was delicious and the talks great (even though I already had enough of social interactions by this time).
Day #1
After years I decided to skip the keynote (and I still don’t have a feeling I missed anything major) and rather ventured between sponsors looking for what might be interesting for me and my customers. I skipped a lot of booth immediately as most of them had some version of „we do AI“ on them and I already had enough of this.
But there were plenty of great things I saw:
Copado and their Org Intelligence – this is high on my list of things to test as it is free (up to 50 queries per month) and looks like it can answer a lot of questions the admins might have and would be hard to find in the documentation of their org (doesn’t matter whether it in fact exists or not);
Gearset and Sweep have potentially the very same capabilities and I need to check those as well;
ChargeBee CPQ – as I haven’t heard the best feedback on the new Revenue Cloud (still want to try it somewhere on my own), this product looks really good, is directly integrated with billing and might be just enough for most clients;
Blackthorn solution for events – so far we run CzechDreamin without any specializied app (if I skip Eventbrite) but we might want to try this, where all the data are at one place;
GridMate was in my head always a shortcut for better list views, where I can edit data but also show them grouped and with conditional formatting. But I found out they do way more including Gantt Chart (which I just need for one client) and 30 other components which might be useful;
I played with Hubble just before Dreamforce and need to spent more time with that. The best part for me is the indication whether we have the last version of package installed or not, because keeping on top of this is super hard for me as a lot of ISVs don’t even notify the admins that there is a new version to install;
PDF Butler didn’t disappoint and their new contract management functionality looks really good;
I have on my list of things to learn/test also the SPIFF tool from Salesforce to calculate commissions and I’ve been really surprised how many such tools have been between sponsors – Xactly, Everstage or CaptivateIQ – looks like I need to check more tools and compare them;
Data Cloud 10 tips from implementation wasn’t anything surprising and I would say they apply to all projects – understand what is the product about (this isn’t ESB/ETL while it can do a lot of that; understand data model & capabilities; involve stakeholders soon and often; have clear objetives; start with MVP and scale up; bring only needed data – I was never sure about this one, glad to have it confirmed from another person; think whether business really needs streaming data; use OOTB connectors; don’t forget that credits are counted from processed rows not returned rows and that data ingest from SF is now for free).
Anyone can become an architect but not everyone can be an architect
What a great statement and the way how to become an architect is to get more involved in things outside of your main area – ask for additional responsibilites, invest your time. Don’t be afraid to use AI, ask it questions and let it explain things to you, become a great builder, learn the clouds and know more about many areas – thank you Steve for the ideas! BTW – isn’t it weird that we don’t trust code which juniors generate with AI but we trust the same code when it has been generated by senior developer?
Day #2
Nonprofit breakfast? Most likely, was interesting to see that nonprofit community groups (and developer groups are pretty much the same) count for 10 % of the total number of community groups. Also celebration of 10 years of Community Sprints (if I’m not mistaken, it is written on the shirt I put somewhere).
Developer keynotes and Agentforce Vibes – but how to vibecode to production? That’s the crucial question I see again and again and where the low code platform doesn’t really help much, as users and admins can get really far with their needs and then have the last small request which means you need to redo everything!
Is it Agentforce 360 or 36O? The letter O at the end is so obvious when you hear it for the first time.
How to design UX for something we don’t know would be the usecase? That’s the crucial question today with the agents and the answer is conversational UX but not in the form of the small box we are used to now on all those websites but fullpage view. But where to put it? You guessed right – Slack!
Scale Center and ApexGuru available now for all customer, another tools I probably should check.
30 000 customers already run on the new Cosmos UI which is pretty incredible to me and one of the numbers I don’t really understand – Salesforce still reports 150 000+ customers for last X years, I didn’t really expect that a lot of existing customers would switch and having 30K new customers would be reason to report, right?
Also concert day? And that arrogant guy I will remember for a while!
Day #3
I took a trip to „SalesforceBen HQ“ for their breakfast talk about blended teams – something I see very often and which they didn’t really bring to the edge. Internal blended team (developers, admins and consultants) are definitelly challenge in what people want to do and where they feel should be handovers (don’t overcomplicate it!) but I see the same with customers whom you cannot really control but would wish for the same approach to documentation, deployment, testing, etc.
75% of admins also use coding tools, being it „just“ SOQL or all the way to Visual Studio Code. The new „vibing“ approach looks fantastic, but doing the right thing (no matter whether vibe or not) should be more important
Final advice – slow down to speed up -> take your time to absorb the learning from Dreamforce or any other event, so you can properly utilize them later.
I’ve probably never been to Marketing keynote and it was super interesting. I just wish to really meet a team who would be able to implement all of that!
Conversational marketing, where you can really reply to the email (good bye „noreply@“ addresses) and you’ll get reply back. No wonder the email should have „This email can talk back, hit reply“ somewhere visibly as we aren’t used to this approach. The whole ping-pong about scheduling a meeting and later rescheduling was fantastic and I wonder how much it takes to prepare.
Which was actually the next thing they showed – so far it takes about 8 weeks to prepare marketing campaign. The whole journey, branding, wording, emails, etc. With Agentforce you can do it in minutes, but would the management be really ok with the emails generated by AI if now they are happy to fight over every single word? Would it sell enough and whom to blame if not? I understand it can be real helper, but doubt we will be brave enough to let it run the whole campaign on its own.
BTW: why they still happily use the „ExactTarget“ reference when speaking about specific cloud, if they bought it 12 years ago? (I’ll skip why it still looks the same)
What is also slightly confusing to me is the new Marketing Cloud Next/Growth/Advanced and how the flow is involved. It was great to see how they redid the campaign layout, but I still struggle when to use Journey builder and when to go with flow and why flow should decide which journey will be triggered it this can be decided in the journey.
Which is nice seque to the presentation about 5 learning from real implementation of MC Next, the most crucial (to me) was that admins need to be involved and marketers should not be the one who will design the flow structure.
Day #4
Go home! It was a long week, so last day in San Francisco, bridge, Painted Ladies, Waymo and back to Europe! Try to catch up with all the news and bits on social networks and there Leanne is the winner with her own summary of Dreamforce – what a week!
Flight to San Francisco, what a better time to read through some short book such as release notes (955 pages this time). For a second I played with Agentforce, which is now running on Salesforce Help site, whether it can provide me a summary tailored for our instance, but got nothing useful out of that. Back to manual reading, which I still prefer to al shortcuts (which I love to use elsewhere). Worth to say that Chris Pearson has been more successful and the result looks awesome!
Here is the list of interesting things we got over the weekend in our instances:
shorten certificate validation – surprisingly I didn’t notice any help with auto generating new besides the possibility to use relevant API. And it is quite annoying generate the certificate for SSO once a year, if we have to do it more often now … :_(
static reference lines in dashboards – I somehow didn’t realize it isn’t there yet plus why – if I remember correcly – you can have them dynamic in reports
translate reports in dashboards – never noticed it isn’t possible
you can type field name you want to add into list view – but it isn’t obvious at all as there isn’t search box above
Lightning Out 2.0 – while Lightning Out never made it out of beta, this is the future now, supporting publishing LWC components out of Salesforce with strong security and performance
SLDS 2 (aka Cosmos) is GA with Dark Mode in beta – need to be enabled by admin
ApexDoc a new standard for documenting your apex code
External Services can handle larger payloads (as I read it then Apex) without hitting a heap limit. Well, we still speak about 16MB only
Named Query API looks interesting as it should be more efficient than flow/apex
via Salesforce DX you can change based on which org a sandbox is refreshed – that might be handy time to time
new Event Studio app will allow you to monitor Platform Event consumption/publishing
when using Lightning Dualbox in your components you can search in it – similar to the way when adding field to list view (see above)
topics in Agentforce has been updated with plenty of new actions, which is pretty interesting as it also means that the users need to learn the new ways they can use agents in
legacy chat is being replaced by Enhanced Chat (formerly Messaging for In-App and Web)
better capture of data on forms in Marketing Cloud Next – no more JavaScript which would set the field based on URL parameters
support of sandboxes in Marketing Cloud Next
Einstein Summary component (if you have Agentforce for Sales) looks really cool
threaded view in Activity timeline when using Einstein Activity Capture
store emails captured by EAC in Salesforce
decision in flow can be done by AI – just write in plain text what when
searching for variables in a flow – finally I don’t need to think about which loop variable is the right one plus plenty of other filtering down
triggering a flow on attaching a file to a record – wow!
previewing screen flow in an Experience Cloud site – no need to activate it and check on the site, can see it directly in the debugger. I would so use it a month ago!
compare flow versions – yes, we have git, but seeing it right in the Salesforce is still super beneficial
persistent logging for flows and reporting how they are used – this looks super promising for some ROI calculation
approval flow debug – it was basically impossible to debug approval process, now with the move to flows we will be also able to debug it, which is super cool
Salesforce Channels in Slacks are automatically created when you start typing into the Slack component on a record which doesn’t have channel already – something I need to dig deeper and see the benefits, as my customers typically didn’t integrate Slack & Salesforce yet and just posting links and have a chat about them
database encryption – I didn’t realize that Salesforce changed the underlying database (no Oracle anymore on Hyperforce) and it looks like you are able to encrypt it with your own keys even without purchasing Shield (I would say the release notes aren’t exactly clear here as in one place it speaks about the Shield, in another one it looks like it is available for everyone) but Shield is required
guidance on how to close a case based on historical cases – part of Agentforce for Service, but feel it can really be beneficial. Actually at the same time, such information should be probably part of knowledge base and the agents should not really based it on previous cases
case descriptions with rich formatting (beta) – I somehow had a feeling it is already possible but obviously not
maximum of 250 emails per 24 hours per case, no limit previously, but even this number looks ridiculously high
Einstein for Service allows you to use standardized Email Prompt Templates for typical use-cases – after becoming the Agentblazer I can feel this makes a lot of sense
Time to Next Action in case list views when using Entitlements – finally!
Not bad but somehow I expected a bit more from almost 1000 pages. Obviously there are plenty of updates to all those different industry clouds I don’t really pay much attention or nothing I would consider major in Education, Nonprofit, and Manufacturing. Revenue Cloud including Billing got a lot of improvements but somehow most of them I considered as expected and was rather surprised it wasn’t there already.
Tentokrát mi pomohl Pepa a povedlo se mu do podcastu přivézt Michala z Anywhere. S ním jsme si povídali o testování.
Testování, taková ta popelka, která se nějak zašmudlá, aby to vypadalo a pak všichni budou doufat. Včetně klienta, ani ti jeho lidé typicky netestují lépe, protože život stejně vždycky ukáže.
Michal rád káže dobro, takže nám kázal i tady. A budu rád i za zpětnou vazbu na zvuk, já moc neslyším tak je mi to jedno, ale tentokrát jsme se sešli v profi studiu, takže pokud vám to zní fakt peckově tak dejte vědět, třeba si ten zážitek zopakuji u dalšího podcastu.
Zápisky
v Salesforce světě si lidi často myslí, že testování = code coverage, ale to s kvalitou nemá nic společného, zapojte byznys ať si i vývojář je jistý, že daný test dává smysl a opravdu k něčemu je
testování ve Salesforce nesmí končit na 75 %, ale začínat u mindsetu
ideál je, aby tester byl u zrodu feature – ne až na konci, když už je hotovo
business by neměl být ‘poslední kontrolor’, ale ‘spolutvůrce akceptačních kritérií’
testing pyramidu všichni znají, ale málokdo se jí řídí, unit testů by mělo být nejvíc, všechno otestují a pak už se to jenom skládá. Když člověk dělá jenom end-to-end testy nebo smoke testy tak se mu pak těžko hledá proč to nefunguje a kde je zakopaný pes
dnes už tester ≠ “klikač”, ale spíš “junior developer / software engineer”, ideálně se schopností psát skripty a rozumět businessu
tester je často jediný člověk v týmu, který se dívá na produkt očima uživatele — a to je strašně podhodnocená role
technická kvalita bez byznysové relevance je k ničemu. A naopak byznysová logika bez technické stability taky
největší dopad na kvalitu nemá počet testů, ale to, jak brzo odhalíme chybu
na frontend (LWC / UI) skoro nikdo netestuje – všichni spoléhají na manuální UAT
když chceš, aby lidi testovali, musíš jim to udělat jednoduché — automatizace nesmí být raketová věda
většina bugů se neobjeví v happy path, ale v momentě, kdy uživatel udělá něco, co jsme nepředpokládali
rozdělení testování na backend (Apex) a frontend (LWC / UI automatizace) – v Salesforce je frontend testing často zanedbaný, vše se spoléhá jen na Apex testy
governance a Center of Excellence
automatizace vs. manuální testování – manuál má pořád své místo, ale cílem je co nejvíc automatizovat
AI ti testy sama nenapíše, ale když jí dáš správný kontext, umí ti ušetřit hromadu času
testovat AI? To bude legrace a uvidíme jak na to (popravdě příslušný Trailhead modul jsem si užil, byť mám pocit, že výsledek těch testů v SF je takový trochu jiný než bych chtěl
It has been a while since Salesforce prepared the Agentblazer ranks and over the summer they publish the last one – Legend. It took me a while to pass those 20 hours of fantastic learning, a handful of superbadges (which I feel went really down in their toughness and are easy to pass), some videos and plenty of other learning.
Now I should be the expert and lightning torch for Agentforce implementation. Sadly I’m still not 100% convinced I do have the customers for this technology.
The benefits
Let’s start with the benefits and ideal customer. If you already have an Experience Cloud site for your customers and partners, it makes a lot of sense to extend it with Agentforce. The conversational interface should be awesome for them and way better than trying to navigate the system and find what they are looking for.
Bonus points if you have knowledge base. Not necessarily the Salesforce KnowledgeBase, it is absolutely fine if you have just a documents with all the text and everything which Agentforce can index, act on it or takes answers from it. Extend the Experience Cloud with agents is pretty straightforward and you have a lot to offer to customers.
If you don’t have anything external looking and if you don’t have any written knowledge to share it is way harder to find some great use-cases. You can obviously provide it to internal users as a tool to summarize records and do some automated actions, but as a user I would prefer to click a button rather than having a dialog with some tool.
why would internal people constantly write to an agent rather than clicking a button?
The Prompt Templates are – from my point of view – perfect use-case for internal people. While I originally hated the idea that every single email will have different wording and I need to check what it generated before sending it to customer, now I can see the benefits. I would still hesitate to use it for „transactional“ emails, such as confirmation of order, thank you email for your donation or things like that, but for all those „soft“ email as „Hi Martin, long time no see, what happened after our last meeting about opportunity XYZ“ I can see how it can save the work for users. And having an Apex action calling prompt template to summarize account info all of that in a button I can click = wonderful!
Also the whole SDR topic, something we cover before with marketing automation/journey builder/cadence builder can be automated with Agentforce but again it means that you have the process described somewhere. If you just hope the people will figure out how to do their jobs you won’t be able to use Agentforce.
I’m absolutely amazed by the language processing capabilities – in bots we had to prepare some placeholders, in email parsing we had to rely on specific anchors to get the right values. Here you just say „the user need to provide their email and customer number“ and the user can reply with any sentence, any order, don’t use specific words and it will correctly take the right values.
The lows
We should also speak about the lows or drawbacks. Agentforce is organized around Topics, Actions and Instructions and it doesn’t give you a blank sheet to do anything. Meaning if you want to integrate with some external system you need to create such an action.
If you want it to do anything in the system you need to create a Topic and Instruction for it and add at least some default Actions.
The interesting/good part about Instructions is that they can be quiet general and the system gets the meaning of it and even though users are using different words it recognize it is the same intent.
Admin point of view
Speaking about admin point of view that’s the major change where enablement needs to be strong.
Historically admins (and consultants) didn’t really have to understand the business process, it was enough when the business told us „do XYZ when ABC“. Obviously the more you understood why the better the solution, but we could survive.
With Agentforce you need to really understand what is behind as you need to:
prepare topics, which group actions together
write scope what each topic is about, how it should behave, what tone it should use, where are the boundaries, etc.
prepare the actions (read flows) which should be as granular as possible so the agents can combine them together – forget about huge flows doing the whole action as we are used now – screen to get data, find related records, do updates, etc -> it all has to be in separated flows so agents can reuse them. This will lead to hundreds of flows (naming structure!) and admins will need to think a lot where to make the cut.
write specific instructions – if you want this you need to have this info and call this flow then this flow, if they want this you need to reply in this way. If you don’t really understand the business process you have no chance to write proper instructions. Business can still provide them, but …
Tomas Cupr wrote a great article about the change in AI and how we are/should be moving from simple instructions to more complex assignments to the agents, setting the whole context and all info we have available. And that’s precisely it – simple sentence as an instruction won’t really make it, you need to explain everything to it as you would do to junior. Easy for business, hard for admins who are missing the context as well.
Licenses?
I feel this wasn’t covered in the learning at all. I have no idea what you need to buy or how much it will cost. I know the Salesforce Foundation is here, which you can enable for free and it will give you some kind of access to it but besides that I’m lost.
3rd party apps? Oh my, there are so many of them and maybe they are better suited/easier to deploy/cheaper. Propose the right solution in this area is tough as there are hundreds of them, each with their own capabilities perfectly suited for their purpose, so it is all about the company’s vision where to go and how to handle it. The best CTA topic of build versus buy which is hard to decide especially at the beginning of some trend/project. Realm or Qualified are just two examples I saw this week
Go or no-go?
The choice is obviously yours. I can see the power of Prompt Templates as a starter and I’m curious about all the use-cases I’ll see for Agentforce. At the same time I would slow you down with implementing it just because it looks fancy. Proper planning is a must and you need to involve all the relevant stakeholders, which needs some extra thinking as the group can be wider than you originally thought.
