FROM: A10X FIELD DESKTO: FORWARD DEPLOYED ENGINEERSRE: BRIEF №5 · THREE MOVESDATE: AUGUST 2026 · WK 34
Brief №5. Three moves as always: what changed, why it matters to an FDE, and what to do this week. No link dumps. Two numbers went around the industry this week. One of them is sixteen months out of date, and the other one has no published denominator. Both are the kind of thing you get asked about in a customer meeting.
DeploymentsThe EU high-risk deadline everyone planned around moved by sixteen months, six days before it would have landed
What changed. The Digital Omnibus on AI (Regulation (EU) 2026/1744) was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It defers the AI Act's high-risk obligations for standalone Annex III systems from 2 August 2026 to 2 December 2027, and for AI embedded in regulated products under Annex I from 2 August 2027 to 2 August 2028. The Commission first proposed the deferral on 19 November 2025, and it was still described as expected rather than adopted as late as this summer, which is why a lot of material written in good faith now says the wrong thing. What was not deferred is the part people skip. The duty to tell a person they are interacting with an AI system, under Article 50, applied from 2 August 2026. The separate duty to mark AI-generated content in machine-readable form carries a four month transition for systems already on the market, so that one bites on 2 December 2026. The general purpose AI obligations and the Article 5 prohibitions were already live well before any of this, the AI literacy provisions in Article 4 survived but were reworded, and the Omnibus adds two new prohibitions, on non-consensual intimate imagery and CSAM, from 2 December 2026.
Why it matters to an FDE. Search this and you will still find recent, confident write-ups saying high-risk obligations became enforceable on 2 August 2026. That is wrong by sixteen months, and it is wrong in the direction that costs money: if you scoped a customer's agent rollout around getting conformity assessment done by August, you scoped urgency that does not exist and probably traded away something that did. The reverse error is worse though. The transparency obligation is live right now, and it is not a policy document, it is a build item. An agent that talks to a person has to disclose that it is an agent. That is a string, in a repo, on a surface, and it is the FDE who either shipped it or did not. This is the same lesson as last week from a different direction: the model was never the hard part, the boundary between the system and the people using it is.
→ DO THIS WEEK: take the one agent you have in front of a customer's users and answer two questions from the code rather than from the deck. Does it disclose that it is an AI at the start of a conversation, and what file is that string in? Then open whatever plan or deck your team is using, find the compliance date on it, and check it against the Official Journal date above. If it says August 2026, you have just found the highest-value thing you will say in your next internal meeting.Benchmarks"Agent deployments nearly tripled" is everywhere this week, and nobody has published what it is a rate of
What changed. On 10 August Salesforce published the second edition of its Agentic Enterprise Index, and the numbers moved fast: organizations nearly tripled their count of active agents, average agent creation time fell 53 percent, and the average agent went from roughly 2 skills to 6, reaching about 9 in retail at peak. Salesforce's own headline is the more conservative more than double year over year; most of the coverage rounded it up to nearly triple. The index is built on aggregated product usage from Agentforce rather than on a survey.
Why it matters to an FDE. Two things worth noticing, and the second one is the useful one. First, this is vendor telemetry from the vendor's own agent product. That is not a reason to dismiss it, it is a reason to say so out loud when you quote it. Second: the coverage does not disclose the sample. Independent analyst write-ups of this same index note explicitly that the organizations included, the industry composition, the sample size and the geographic scope are not given. So the number is a real measurement of something, and you cannot say what it is a rate of. That matters because of exactly how it will get used. Somebody will put it in front of your customer as a reason their rollout is behind. A growth multiple measured across an unknown set of companies who were already running agents on one vendor's platform tells you nothing about whether this deployment is on track. Getting comfortable saying "I don't know what that is a percentage of" in a room full of executives is a genuine FDE skill, and it is rarer than it sounds.
→ DO THIS WEEK: pick the single benchmark you quote most often, the one you reach for without checking, and try to find its denominator in one click. Who was in the sample, how many, over what period. If you cannot find it in a minute, either stop quoting the number or start quoting it with the sample attached. Both are fine. Quoting it bare is the thing to stop doing.HiringTwenty-two forward deployed requisitions, all still open, checked against the employers' APIs rather than their job pages
What changed. We re-checked every role on the A10X board against each employer's own applicant tracking API on 15 August 2026: 22 roles, 22 still live and listed, 0 dead, across 14 boards. The reason we check that way rather than clicking the apply links is that clicking them does not work. A requisition survives a rename, so the apply URL keeps returning a healthy page long after the role has been retitled or filled. Worse, the three big applicant tracking systems disagree about how they fail: one returns a client-rendered shell that is roughly the same size as a real posting, one serves you the company's whole job index instead, and only one returns an honest 404. A sweep of twenty-two links can come back twenty-two for twenty-two while a role is dead.
Why it matters to an FDE. This is the single cheapest edge available to anyone applying for these roles right now, and almost nobody uses it. The posting APIs behind Ashby, Greenhouse and Lever are public and unauthenticated. They will tell you whether a role is actually listed, when it was published, and frequently the full description text and the compensation band, including bands that are not rendered on the page you were reading. Two uses. If you are hiring, it is how you stop your careers page lying to candidates. If you are applying, it is how you stop spending an evening on a requisition that has been closed for five weeks, and it is how you walk into the screen already knowing the band. It is also, incidentally, the most FDE thing in this brief: the data was always there in a structured form, and the interface everyone uses hides it.
→ DO THIS WEEK: take one job you are about to spend a real evening on. Find the employer's board slug in the apply URL, call the posting API for it, and read the row yourself. Check isListed before you write a line of the application. If the role is on our board, we have already done this for you as of 15 August, but do it once by hand anyway. It takes four minutes and you will not go back to trusting a careers page.The through-line
Both of this week's headline numbers traveled further than their evidence did. The compliance date is quoted by people who did not check whether it moved, and the deployment multiple is quoted by people who cannot say who is in the sample. The job in the room is not to know more numbers. It is to be the person who asks which date this is, and what this is a percentage of, and then goes and reads the source. That habit is most of what separates an engineer a customer trusts from one they merely tolerate.
See you next Monday.
— THE A10X FIELD DESK
Sources: EUR-Lex, Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal, 24 July 2026; European Commission, AI Omnibus enters into force, 27 July 2026; European Commission, Regulatory framework for AI; Salesforce, Agentic Enterprise Index: insights 2026; Futurum Group, Salesforce's Agentic Enterprise Index: a paradigm shift in AI deployment.