A rather 'unsexy' thing recently appeared on the OCA community blog: someone labeled each of the 687 features in Odoo 20's official release notes one by one as 'available in Community edition', 'partially available', or 'Enterprise-only', and attached evidence to each item.
The person who did this is Jan-Marten Veddeler (Humanilog); at the end of the page he deliberately wrote two lines: This is not an official publication of Odoo S.A.; this is an early draft, 267 items directly verified, 362 items only verified at module level, and 58 items still to be reviewed; there are certainly errors, and corrections are welcome.
I stared at “58 items pending review” for a while. Over the past two years, when people talk about enterprise AI, the conversation has basically centered on models, agents, and autonomous execution; but when it actually comes to the delivery front lines, the most laborious and least desirable work is often exactly this kind of task: checking boundaries one by one to make them clear. This checklist and the article about FDE published in Economic Daily on November 8 are actually two sides of the same thing.
I. A checklist that turns “Is there one?” into “Which version, which layer, and who verified it?”
Let's put the numbers out first.
| Category | Quantity | Proportion | Instructions |
|---|---|---|---|
| Community Edition available | 341 items | 50% | Attach an actual screenshot of the community edition database as evidence |
| Partially available | 90 items | 13% | The entry specifies which part requires the Enterprise Edition |
| Enterprise Edition Exclusive | 256 items | 37% | Most come from applications and industry packages that exist only in the Enterprise Edition |
What's really interesting is the distribution, not the total: the apps used most in daily work are basically all in the Community edition—Website 55/60, General 36/41, eCommerce 34/38, Sales 23/27, Manufacturing 20/24, Inventory 19/22.
That 37% of Enterprise Edition mainly comes from apps that exist only in Enterprise Edition: AI 20 items, Planning 20, Appointments 19, Payroll 15, Sign 15, plus 25 industry packages. In other words, if a customer points to the official release notes and asks, “Can we use Odoo 20’s AI?”, the answer isn’t in the feature list; it’s in which license they bought. Spreadsheet is even more typical: all 30 items are marked “partial”: the spreadsheet engine is in Community Edition and even supports Community Edition dashboards, but to create and edit spreadsheets, you need Enterprise Edition.
| Columns in the list | the things found out | Impact on delivery |
|---|---|---|
| Déjà vu | 16 "new features" were already available in OCA modules years ago: partner_identification (8.0), website_breadcrumb (8.0), sale_order_line_sequence (9.0), website_llms (16.0), etc. | Don't rush to uninstall existing modules—similar core functionality does not necessarily cover all the capabilities of OCA modules |
| Removed from CE | Turkey Nilvera e-invoice/e-delivery note moved to Enterprise edition; Mondial Relay community connector removed, switched to Enterprise Sendcloud; project profitability panel removed, budget analysis moved to Enterprise edition | The three items each have OCA alternatives (delivery_sendcloud_oca, account_budget_oca), and the upgrade plan needs to be rewritten. |
| For developers | A single ir.access model replaces ir.model.access and ir.rule; OWL 3 replaces OWL 2; Material Symbols replaces Font Awesome; 49 Community Edition modules were removed or merged | Every self-developed module with a security file must be gone through once. |
| Evidence level | 267 items directly verified / 362 module-level items / 58 items pending review | Conclusions can be questioned, but each one can be traced back to evidence |
The real value of this checklist lies not in whether its "conclusions are correct," but in how it breaks a vague "supported" into four fields that can be questioned: version, license layer, enablement steps, and evidence level. Anyone who has done acceptance testing knows that these four fields are what a contract annex should look like.
We have seen too many launch reports that say 'the AI feature is enabled,' paired with a chat screenshot. When a real problem occurs, no one can clearly say which configuration it was using at the time, which records it accessed, or whether it stopped at a point where a human decision was needed. The checklist is not bureaucracy; it is the only thing that allows you to trace back what happened when something goes wrong.
II. Upgrading is not “it passes as long as it opens”
The trickier part is in the developer section, and most of this is not in Odoo's official release notes.
- The permission model has changed its underlying implementation. A single ir.access model replaces ir.model.access and ir.rule, and modules now use security/ir.access.csv. All modules with security files are affected; simply renaming a file so the system can load it does not mean the authorization combinations still carry their original meaning.
- The tech stack changes in tandem. OWL 3 replaces OWL 2 (with a compatibility layer); Material Symbols replaces Font Awesome, so you have to go through all the fa fa-* icons; in accounting, parent accounts replace account.group, account codes become optional, and payment status is renamed.
- The modules have been merged. stock_picking_batch merged into stock, website_sale_wishlist and website_sale_comparison merged into website_sale—OCA modules that depend on them need to change depends.
- There are also naming collisions. Odoo 20 Community Edition added mail_tracking and mail_tracking_mass_mailing, whose technical names are exactly the same as those of existing OCA/mail modules, so these OCA modules cannot be migrated to 20.0 under their original names.
The official version provides automatic code rewriting, and the page recommends running it on the repository first:
odoo-bin upgrade_code --addons-path=<repo> --from 19.0 --dry-run
Note that it is --dry-run. The biggest taboo in this kind of thing is 'just run it and see'—the database can open, the menu can be clicked, and the agent can even answer a couple of familiar questions, but none of that is enough to prove that the original configuration mapping is correct, that the request and billing paths are correct, or that it will stop when it should. The page also specifically includes a Needs IAP filter to single out items that depend on paid services, plus five tabs for implementers to filter as needed.
These are easy pitfalls here: mistaking "the Community Edition has dashboards" for "the Community Edition has tables," mistaking "AI applications can be installed" for "AI features are included in the license," and mistaking "the upgrade script didn’t report errors" for "the permission policy hasn’t been broadened." None of these three things show up in the demo environment; they only all surface together at month-end closing and audit.
III. People who can get on site have become the most expensive resource
Someone has to verify the list, someone has to guard the boundaries, and someone has to redo the processes. The Economic Daily article on November 8, “Competency Benchmarks for Forward Deployed Engineers Are Accelerating into Shape,” is exactly about this group of people beginning to be formally priced.
According to statistics from the Liepin Big Data Research Institute, from September last year to August this year, the supply-demand ratio for domestic FDE (Forward Deployed Engineer) talent was only 0.28—four positions competing for one person. On the policy side, the Ministry of Industry and Information Technology has included FDE in a national-level policy document; Shanghai has proposed building a reserve pool of more than 1,000 people, and Wuhan Optics Valley plans to cultivate 1,000 in three years.
The most concrete step on the industry side is the FDE training and certification jointly released by Lenovo and the Electronic Industry Standardization Institute of the Ministry of Industry and Information Technology—the industry's first certification jointly recognized by an enterprise and a professional industry talent standardization body. The courses are divided into three major series: enterprise AI design methods, agent workshops and basic AI capabilities, and project delivery implementation and operations, with a total investment of approximately 120 hours, most of which is devoted to after-class exercises and integrated comprehensive practical training.
The most worthwhile sentence to copy into your own SOW is this exam requirement: The practical exam explicitly requires completing designated tasks in a preconfigured environment and submitting evidence of the process and results.
The test isn't whether you can talk about it, but whether you can get the work done in a given environment and leave behind a process that can be reviewed.
Lenovo's definition of FDE capabilities is not tied to tools either: translating customers' vague business demands into 'what to do, what not to do, and who is responsible,' and translating technical metrics back into business language. Behind this is its own practice—internally, 60% of business processes have already been taken over by agents, and externally it has accumulated more than 200 verifiable and deliverable agent solutions, with 128 in the manufacturing industry alone.
Xiong Meng, Party Secretary of the China Federation of Industrial Economics, offered a fairly accurate dividing line: "The only dividing line for identifying an FDE is whether field experience has been systematically consolidated into platform capabilities." This aligns with another survey on November 8—in much. Consulting's "Odoo News November 2026," 86% of 150 surveyed manufacturing enterprises listed "lack of internal resources" as the biggest obstacle to introducing or modernizing ERP. The former was talking about the problem; the latter says someone has begun setting the yardstick for solutions.
Tencent's FDE co-creation on the automaker side takes another form: it provides technical people, while the automaker provides business people to form an agile team and pick high-value scenarios; after a certain OEM used it in the product definition and validation stages, overall efficiency increased by 90%, and production-sales coordination decisions were compressed from two to three weeks or even a month to one or two days. The numbers sound impressive, but what really determines whether it can be replicated is that on-site division of labor: 'who raises requirements, who judges whether it is worth it, and who accepts it'.
Domestic FDE talent supply-demand ratio (Liepin Big Data Research Institute, 2025-09 to 2026-08)—equivalent to four positions competing for one person
Looking at the OCA checklist together with this FDE certification, the structure is symmetrical: one turns system boundaries into verifiable items, and the other turns human capabilities into assessable evidence. The phrase in between, “submit process and result evidence,” uses the same logic on both sides.
IV. Three Recommendations for Implementers and CIOs
- Write acceptance criteria as line items, not as adjectives. In the SOW, list the version, license tier, enablement steps, acceptance test cases, and evidence format item by item; replace terms such as "supported," "live," and "good results" with reproducible test cases. Directly borrow the line from the FDE hands-on exam: submit process and result evidence.
- Run the checklist before upgrading, then talk about AI. Review your own modules across the five columns Highlights / All items / Déjà vu / Removed from CE / For developers, focusing on modules that have security files, depend on merged modules, or have technical-name collisions with OCA; first run
upgrade_code --dry-run, and do a separate regression pass for AI workflows—don't use “the interface opens” as the pass criterion. - Turn field experience into assets. Every project leaves three things behind: a configuration mapping table, an acceptance ledger, and a fault retrospective. When choosing partners, ask less “how many similar projects have you done” and more “what does the previous client’s acceptance ledger look like—can you show me one redacted page?” Capability that lives in individuals leaves when the project ends; capability that lives in the ledger stays.
Models will change generations, and the feature list will be rewritten once every autumn. But “who checks the boundaries clearly” and “who can get on site and finish the work” — these two things will not automatically be in place. The delivery quality of AI projects ultimately still rests on people and ledgers — this work isn’t sexy, but it determines whether those pretty demos up front can actually pass month-end closing.
Sources: the Odoo Community Association community blog “Odoo 20: what is actually in Community?” and the smaddlsoft/odoo-community-release-notes repository (an unofficial draft; the data files are still being updated as of 2026-11-03), Economic Daily’s “Forward Deployed Engineer Competency Benchmark Accelerates Its Formation” (2026-11-08), Lenovo × the Ministry of Industry and Information Technology Electronics Industry Standardization Research Institute FDE training and certification, much. Consulting’s “Odoo News November 2026”, and Jiemian News · Lingjing / NetEase (Tencent Smart Mobility FDE co-creation, 11-08). The efficiency figures in this article reflect the vendors’ or institutions’ own stated metrics.
Zhongdi Information (Zodioo) · Enterprise Digitalization Insights · 2026-11-09
